a.txt

A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O


siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


Bui	Bui	NNP	B-NP	O
Thang	Thang	NNP	I-NP	O
Ngoc	Ngoc	NNP	I-NP	O


Japan	Japan	NNP	B-NP	O
Advanced	Advanced	NNP	I-NP	O
Institute	Institute	NNP	I-NP	O
of	of	IN	B-PP	O
Science	Science	NNP	B-NP	O
and	and	CC	O	O
Technology	Technology	NNP	B-NP	O

1-1	1-1	CD	B-NP	O
Asahidai	Asahidai	NNP	I-NP	O
,	,	,	O	O
Nomi	Nomi	NNP	B-NP	O
City	City	NNP	I-NP	O
,	,	,	O	O
Ishikawa	Ishikawa	NNP	B-NP	O
,	,	,	O	O
923-1211	923-1211	CD	B-NP	O
Japan	Japan	NNP	B-NP	O


thangbn	thangbn	IN	B-PP	O
@	@	NN	B-NP	O
jaist.ac.jp	jaist.ac.jp	NN	I-NP	O


Abstract.	Abstract.	NNP	B-NP	O
Short	Short	NNP	I-NP	O
interfering	interfere	VBG	B-VP	O
RNA	RNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
(	(	(	O	O
siRNAs	siRNA	NNS	B-NP	O
)	)	)	O	O
can	can	MD	B-VP	O
knockdown	knockdown	VB	I-VP	O

target	target	NN	B-NP	B-DNA
genes	gene	NNS	I-NP	I-DNA
and	and	CC	O	O
thus	thus	RB	B-ADVP	O
have	have	VBP	B-VP	O
an	an	DT	B-NP	O
immense	immense	JJ	I-NP	O
impact	impact	NN	I-NP	O
on	on	IN	B-PP	O
biology	biology	NN	B-NP	O
and	and	CC	I-NP	O
pharmacy	pharmacy	NN	I-NP	O

research.	research.	LS	B-LST	O
The	The	DT	B-NP	O
key	key	JJ	I-NP	O
question	question	NN	I-NP	O
of	of	IN	B-PP	O
which	which	WDT	B-NP	O
siRNAs	siRNA	NNS	B-NP	B-RNA
have	have	VBP	B-VP	O
high	high	JJ	B-NP	O
knockdown	knockdown	NN	I-NP	O
ability	ability	NN	I-NP	O

in	in	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
research	research	NN	I-NP	O
remains	remain	VBZ	B-VP	O
challenging	challenging	JJ	B-ADJP	O
as	as	IN	B-PP	O
current	current	JJ	B-NP	O
known	know	VBN	I-NP	O
results	result	NNS	I-NP	O
are	be	VBP	B-VP	O
still	still	RB	B-ADVP	O

far	far	RB	B-ADVP	O
from	from	IN	B-PP	O
expectation.	expectation.	FW	B-ADJP	O
This	This	DT	B-NP	O
work	work	NN	I-NP	O
aims	aim	VBZ	B-VP	O
to	to	TO	I-VP	O
develop	develop	VB	I-VP	O
a	a	DT	B-NP	O
generic	generic	JJ	I-NP	O
framework	framework	NN	I-NP	O
to	to	TO	B-PP	O

enhance	enhance	VB	B-VP	O
siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	NN	I-NP	O
efficacy	efficacy	NN	I-NP	O
prediction.	prediction.	VBP	B-VP	O
The	The	DT	B-NP	O
key	key	JJ	I-NP	O
idea	idea	NN	I-NP	O
is	be	VBZ	B-VP	O
first	first	VBN	I-VP	O
to	to	TO	B-PP	O

enrich	enrich	JJ	B-NP	O
siRNA	siRNA	NN	I-NP	O
sequences	sequence	NNS	I-NP	O
by	by	IN	B-PP	O
incorporating	incorporate	VBG	B-VP	O
them	them	PRP	B-NP	O
with	with	IN	B-PP	O
rules	rule	NNS	B-NP	O
found	find	VBN	B-VP	O
for	for	IN	B-PP	O
de-	de-	NN	B-NP	O

signing	sign	VBG	B-VP	O
effective	effective	JJ	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
and	and	CC	O	O
representing	represent	VBG	B-VP	O
them	them	PRP	B-NP	O
as	as	IN	B-PP	O
transformed	transform	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
,	,	,	O	O

then	then	RB	B-ADVP	O
to	to	TO	B-VP	O
employ	employ	VB	I-VP	O
the	the	DT	B-NP	O
bilinear	bilinear	JJ	I-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O
to	to	TO	B-VP	O
do	do	VB	I-VP	O
prediction	prediction	NN	B-NP	O
on	on	IN	B-PP	O
those	those	DT	B-NP	O

matrices.	matrices.	NN	B-NP	O
Experiments	Experiment	NNS	I-NP	O
show	show	VBP	B-VP	O
that	that	IN	B-SBAR	O
the	the	DT	B-NP	O
proposed	propose	VBN	I-NP	O
method	method	NN	I-NP	O
achieves	achieve	VBZ	B-VP	O
results	result	NNS	B-NP	O

better	well	RBR	B-ADVP	O
than	than	IN	B-PP	O
existing	exist	VBG	B-NP	O
models	model	NNS	I-NP	O
in	in	IN	B-PP	O
most	most	JJS	B-NP	O
cases	case	NNS	I-NP	O
.	.	.	O	O


1	1	CD	B-NP	O


Introduction	Introduction	NN	B-NP	O


In	In	IN	B-PP	O
2006	2006	CD	B-NP	O
,	,	,	O	O
Fire	Fire	NNP	B-NP	O
and	and	CC	I-NP	O
Mello	Mello	NNP	I-NP	O
received	receive	VBD	B-VP	O
their	their	PRP$	B-NP	O
Nobel	Nobel	NNP	I-NP	O
Prize	Prize	NNP	I-NP	O
for	for	IN	B-PP	O
their	their	PRP$	B-NP	O
contributions	contribution	NNS	I-NP	O
to	to	TO	B-PP	O

research	research	NN	B-NP	O
on	on	IN	B-PP	O
RNA	RNA	NN	B-NP	O
interference	interference	NN	I-NP	O
(	(	(	O	O
RNAi	RNAi	NN	B-NP	B-protein
)	)	)	O	O
that	that	WDT	B-NP	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
biological	biological	JJ	I-NP	O
process	process	NN	I-NP	O
in	in	IN	B-PP	O
which	which	WDT	B-NP	O

RNA	RNA	NN	B-NP	B-protein
molecules	molecule	NNS	I-NP	I-protein
inhibit	inhibit	VBP	B-VP	O
gene	gene	NN	B-NP	O
expression	expression	NN	I-NP	O
,	,	,	O	O
typically	typically	RB	B-ADVP	O
by	by	IN	B-PP	O
causing	cause	VBG	B-VP	O
the	the	DT	B-NP	O
destruction	destruction	NN	I-NP	O
of	of	IN	B-PP	O

specific	specific	NN	B-NP	O
mRNA	mRNA	NN	I-NP	O
molecules.	molecules.	NN	I-NP	O
Their	Their	PRP$	B-NP	O
work	work	NN	I-NP	O
and	and	CC	O	O
that	that	DT	O	O
of	of	IN	B-PP	O
others	other	NNS	B-NP	O
on	on	IN	B-PP	O
discovery	discovery	NN	B-NP	O
of	of	IN	B-PP	O
RNAi	RNAi	NN	B-NP	O

have	have	VB	B-VP	O
had	have	VBN	I-VP	O
an	an	DT	B-NP	O
immense	immense	JJ	I-NP	O
impact	impact	NN	I-NP	O
on	on	IN	B-PP	O
biomedical	biomedical	JJ	B-NP	O
research	research	NN	I-NP	O
and	and	CC	O	O
will	will	MD	B-VP	O
most	most	RBS	B-ADVP	O
likely	likely	RB	I-ADVP	O
lead	lead	VBP	B-VP	O

to	to	TO	B-PP	O
novel	novel	JJ	B-NP	O
medical	medical	JJ	I-NP	O
applications.	applications.	NN	I-NP	O
On	On	IN	B-PP	O
RNAi	RNAi	NNP	B-NP	O
research	research	NN	I-NP	O
,	,	,	O	O
designing	design	VBG	B-VP	O
of	of	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	O
(	(	(	O	O
short	short	JJ	O	O

interfering	interfere	VBG	B-NP	B-RNA
RNAs	RNA	NNS	I-NP	I-RNA
)	)	)	O	O
with	with	IN	B-PP	O
high	high	JJ	B-NP	O
efficacy	efficacy	NN	I-NP	O
is	be	VBZ	B-VP	O
one	one	CD	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
most	most	RBS	I-NP	O
crucial	crucial	JJ	I-NP	O
RNAi	RNAi	NNP	I-NP	O
issues	issue	NNS	I-NP	O
.	.	.	O	O

Highly	Highly	RB	B-NP	O
effective	effective	JJ	I-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
can	can	MD	B-VP	O
be	be	VB	I-VP	O
used	use	VBN	I-VP	O
to	to	TO	B-VP	O
design	design	VB	I-VP	O
drugs	drug	NNS	B-NP	O
for	for	IN	B-PP	O
viral-mediated	viral-mediated	JJ	B-NP	O
diseases	disease	NNS	I-NP	O

such	such	JJ	B-PP	O
as	as	IN	I-PP	O
Influenza	Influenza	NN	B-NP	O
A	A	NN	I-NP	O
virus	virus	NN	I-NP	O
,	,	,	O	O
HIV	HIV	NN	B-NP	O
,	,	,	O	O
Hepatitis	Hepatitis	NN	B-NP	O
B	B	NN	I-NP	O
virus	virus	NN	I-NP	O
,	,	,	O	O
RSV	RSV	NN	B-NP	O
viruses	virus	NNS	I-NP	O
,	,	,	O	O
cancer	cancer	NN	B-NP	O
disease	disease	NN	I-NP	O

and	and	CC	O	O
so	so	RB	B-ADVP	O
on.	on.	RB	I-ADVP	O
As	As	IN	B-PP	O
a	a	DT	B-NP	O
result	result	NN	I-NP	O
,	,	,	O	O
siRNA	siRNA	NN	B-NP	O
silencing	silencing	NN	I-NP	O
is	be	VBZ	B-VP	O
considered	consider	VBN	I-VP	O
one	one	CD	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
most	most	RBS	I-NP	O
promising	promising	JJ	I-NP	O

techniques	technique	NNS	B-NP	O
in	in	IN	B-PP	O
future	future	JJ	B-NP	O
therapy.	therapy.	NN	I-NP	O
Finding	Find	VBG	B-VP	O
highly	highly	RB	B-NP	O
effective	effective	JJ	I-NP	O
siRNAs	siRNA	NNS	I-NP	O
among	among	IN	B-PP	O
thousands	thousand	NNS	B-NP	O

of	of	IN	B-PP	O
potential	potential	JJ	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
for	for	IN	B-PP	O
an	an	DT	B-NP	O
mRNA	mRNA	NN	I-NP	B-RNA
remains	remain	VBZ	B-VP	O
a	a	DT	B-NP	O
great	great	JJ	I-NP	O
challenge	challenge	NN	I-NP	O
.	.	.	O	O


Various	Various	JJ	B-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
have	have	VBP	B-VP	O
been	be	VBN	I-VP	O
found	find	VBN	I-VP	O
by	by	IN	B-PP	O
empirical	empirical	JJ	B-NP	O
processes	process	NNS	I-NP	O
since	since	IN	B-PP	O
1998	1998	CD	B-NP	O
.	.	.	O	O

The	The	DT	B-NP	O
first	first	RBS	I-NP	O
rational	rational	JJ	I-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
was	be	VBD	B-VP	O
detected	detect	VBN	I-VP	O
by	by	IN	B-PP	O
Elibalshir	Elibalshir	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
[	[	(	O	O
2	2	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
They	They	PRP	B-NP	O

suggested	suggest	VBD	B-VP	O
that	that	IN	B-SBAR	O
siRNAs	siRNA	NNS	B-NP	O
having	have	VBG	B-VP	O
19–21	19–21	CD	B-NP	O
nt	nt	NN	I-NP	O
(	(	(	O	O
nucleotide	nucleotide	NN	B-NP	O
)	)	)	O	O
in	in	IN	B-PP	O
length	length	NN	B-NP	O
with	with	IN	B-PP	O
2	2	CD	B-NP	O
nt	nt	NN	I-NP	O
over-	over-	NN	I-NP	O

hangs	hang	NNS	B-NP	O
at	at	IN	B-PP	O
3’	3’	NN	B-NP	O
end	end	NN	I-NP	O
can	can	MD	B-VP	O
efficiently	efficiently	RB	I-VP	O
silence	silence	VB	I-VP	O
mRNAs.	mRNAs.	NNP	B-NP	O
Scherer	Scherer	NNP	I-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
reported	report	VBD	B-VP	O
that	that	IN	B-SBAR	O
the	the	DT	B-NP	O

thermodynamic	thermodynamic	JJ	B-NP	O
properties	property	NNS	I-NP	O
(	(	(	O	O
G/C	G/C	NN	B-NP	O
content	content	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
)	)	)	O	O
to	to	TO	B-VP	O
target	target	VB	I-VP	O
specific	specific	NN	B-NP	B-RNA
mRNAs	mRNA	NNS	I-NP	I-RNA

are	be	VBP	O	O
important	important	JJ	B-NP	O
characteristics	characteristic	NNS	I-NP	O
[	[	(	O	O
11	11	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
Soon	Soon	RB	B-ADVP	O
after	after	IN	B-PP	O
these	these	DT	B-NP	O
works	work	NNS	I-NP	O
,	,	,	O	O
many	many	JJ	B-NP	O
rational	rational	JJ	I-NP	O
design	design	NN	I-NP	O

rules	rule	NNS	B-NP	O
for	for	IN	B-PP	O
effective	effective	JJ	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
have	have	VBP	B-VP	O
been	be	VBN	I-VP	O
found	find	VBN	I-VP	O
,	,	,	O	O
typically	typically	RB	B-NP	O
those	those	DT	I-NP	O
in	in	IN	B-PP	O
[	[	(	O	O
10	10	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
15	15	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
1	1	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
4	4	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O

[	[	(	O	O
7	7	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
14	14	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
For	For	IN	B-PP	O
example	example	NN	B-NP	O
,	,	,	O	O
Reynolds	Reynolds	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
[	[	(	O	O
10	10	CD	B-NP	O
]	]	)	O	B-RNA
analyzed	analyze	VBD	B-VP	I-RNA
180	180	CD	B-NP	I-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
and	and	CC	O	O
found	find	VBD	B-VP	O
eight	eight	CD	B-NP	O

criteria	criterion	NNS	B-NP	O
for	for	IN	B-PP	O
improving	improve	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	O
selection	selection	NN	I-NP	O
:	:	:	O	O
(	(	(	B-LST	O
1	1	LS	I-LST	O
)	)	)	O	O
G/C	G/C	NN	B-NP	O
content	content	NN	I-NP	O
30−52	30−52	CD	B-NP	O
%	%	NN	I-NP	O
,	,	,	O	O
(	(	(	B-LST	O
2	2	LS	I-LST	O
)	)	)	O	O
at	at	IN	B-NP	O
least	least	JJS	I-NP	O
3	3	CD	I-NP	O


V.S.	V.S.	NNP	B-NP	O
Tseng	Tseng	NNP	I-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
(	(	(	O	O
Eds.	Eds.	NNP	B-NP	O
)	)	)	O	O
:	:	:	O	O
PAKDD	PAKDD	NN	B-NP	O
2014	2014	CD	I-NP	O
,	,	,	O	O
Part	Part	NNP	B-NP	O
II	II	NNP	I-NP	O
,	,	,	O	O
LNAI	LNAI	NNP	B-NP	O
8444	8444	CD	I-NP	O
,	,	,	O	O
pp.	pp.	NN	B-NP	B-protein
400–412	400–412	CD	I-NP	I-protein
,	,	,	I-NP	O
2014	2014	CD	I-NP	O
.	.	.	O	O

c	c	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O
Springer	Springer	NNP	B-NP	O
International	International	NNP	I-NP	O
Publishing	Publishing	NNP	I-NP	O
Switzerland	Switzerland	NNP	I-NP	O
2014	2014	CD	I-NP	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


401	401	CD	B-NP	O


As	As	IN	B-SBAR	O
or	or	CC	O	O
Us	U	NNS	B-NP	O
at	at	IN	B-PP	O
positions	position	NNS	B-NP	O
from	from	IN	B-PP	O
15	15	CD	B-NP	O
to	to	TO	B-PP	O
19	19	CD	B-NP	O
,	,	,	O	O
(	(	(	B-LST	O
3	3	LS	I-LST	O
)	)	)	O	O
absence	absence	NN	B-NP	O
of	of	IN	B-PP	O
internal	internal	JJ	B-NP	B-DNA
repeats	repeat	NNS	I-NP	I-DNA
,	,	,	O	O
(	(	(	B-LST	O
4	4	LS	I-LST	O
)	)	)	O	O
an	an	DT	B-NP	O
A	A	NN	I-NP	O
at	at	IN	B-PP	O

position	position	NN	B-NP	B-protein
19	19	CD	I-NP	I-protein
,	,	,	O	O
(	(	(	B-LST	O
5	5	LS	I-LST	O
)	)	)	O	O
an	an	DT	B-NP	O
A	A	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
3	3	CD	I-NP	I-protein
,	,	,	O	O
(	(	(	O	O
6	6	CD	B-NP	O
)	)	)	O	O
a	a	DT	B-NP	O
U	U	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
10	10	CD	I-NP	I-protein
,	,	,	O	O
(	(	(	O	O
7	7	CD	B-NP	O
)	)	)	O	O
a	a	DT	B-NP	O
base	base	NN	I-NP	O
other	other	JJ	B-ADJP	O
than	than	IN	B-SBAR	O

G	G	NN	B-NP	O
or	or	CC	I-NP	O
C	C	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
19	19	CD	I-NP	I-protein
,	,	,	O	O
(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O
a	a	DT	B-NP	O
base	base	NN	I-NP	O
other	other	JJ	B-ADJP	O
than	than	IN	B-PP	O
G	G	NN	B-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
13	13	CD	I-NP	I-protein
.	.	.	O	O


However	However	RB	B-ADVP	O
,	,	,	O	O
most	most	JJS	B-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
design	design	NN	I-NP	O
tools	tool	NNS	I-NP	O
using	use	VBG	B-VP	O
the	the	DT	B-NP	O
above-mentioned	above-mentioned	JJ	I-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O

have	have	VB	B-VP	O
low	low	JJ	B-NP	O
accuracy	accuracy	NN	I-NP	O
,	,	,	O	O
because	because	IN	B-SBAR	O
about	about	IN	B-NP	O
65	65	CD	I-NP	O
%	%	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
predicted	predict	VBN	B-VP	O
as	as	IN	B-PP	O
high	high	JJ	B-NP	O
effective	effective	JJ	I-NP	O

was	be	VBD	B-VP	O
failed	fail	VBN	I-VP	O
when	when	WRB	B-ADVP	O
tested	test	VBN	B-VP	O
experimentally	experimentally	RB	B-ADVP	O
as	as	IN	B-SBAR	O
they	they	PRP	B-NP	O
were	be	VBD	B-VP	O
90	90	CD	B-NP	O
%	%	NN	I-NP	O
in	in	IN	B-PP	O
inhibition	inhibition	NN	B-NP	O
and	and	CC	O	O
near	near	IN	O	O

20	20	CD	B-NP	O
%	%	NN	I-NP	O
of	of	IN	B-PP	O
them	them	PRP	B-NP	O
were	be	VBD	B-VP	O
found	find	VBN	I-VP	O
to	to	TO	I-VP	O
be	be	VB	I-VP	O
inactive	inactive	JJ	B-ADJP	O
[	[	(	O	O
9	9	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
One	One	CD	B-NP	O
reason	reason	NN	I-NP	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
previous	previous	JJ	I-NP	O
empirical	empirical	JJ	I-NP	O

analyses	analysis	NNS	B-NP	O
only	only	RB	B-VP	O
based	base	VBN	I-VP	O
on	on	IN	B-PP	O
small	small	JJ	B-NP	O
datasets	dataset	NNS	I-NP	O
and	and	CC	O	O
focused	focus	VBD	B-VP	O
on	on	IN	B-PP	O
specific	specific	NN	B-NP	O
genes.	genes.	NN	I-NP	O
Therefore	Therefore	RB	B-ADVP	O
,	,	,	O	O

each	each	DT	B-NP	O
of	of	IN	B-PP	O
these	these	DT	B-NP	O
rules	rule	NNS	I-NP	O
certainly	certainly	RB	B-ADVP	O
is	be	VBZ	B-VP	O
poor	poor	JJ	B-ADJP	O
to	to	TO	B-VP	O
individually	individually	RB	I-VP	O
design	design	VB	I-VP	O
effective	effective	JJ	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
.	.	.	O	O


Since	Since	IN	B-PP	O
nearly	nearly	RB	B-NP	O
a	a	DT	I-NP	O
decade	decade	NN	I-NP	O
,	,	,	O	O
machine	machine	NN	B-NP	O
learning	learning	NN	I-NP	O
techniques	technique	NNS	I-NP	O
have	have	VBP	B-VP	O
alternatively	alternatively	RB	I-VP	O
been	be	VBN	I-VP	O

applied	apply	VBN	B-VP	O
to	to	TO	B-VP	O
predict	predict	VB	I-VP	O
knockdown	knockdown	JJ	B-NP	O
efficacy	efficacy	NN	I-NP	O
of	of	IN	B-PP	O
siRNAs.	siRNAs.	NN	B-NP	O
The	The	DT	B-NP	O
first	first	RBS	I-NP	O
predictive	predictive	JJ	I-NP	O
model	model	NN	I-NP	O
was	be	VBD	B-VP	O

proposed	propose	VBN	B-VP	O
by	by	IN	B-PP	O
Huesken	Huesken	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
in	in	IN	B-PP	O
which	which	WDT	B-NP	O
motifs	motif	NNS	B-NP	O
for	for	IN	B-PP	O
effective	effective	JJ	B-NP	O
and	and	CC	I-NP	O
ineffective	ineffective	JJ	I-NP	B-protein
siRNA	siRNA	NN	I-NP	I-protein
se-	se-	NN	I-NP	I-protein

quences	quence	NNS	B-NP	O
were	be	VBD	B-VP	O
detected	detect	VBN	I-VP	O
basing	base	VBG	B-VP	O
on	on	IN	B-PP	O
the	the	DT	B-NP	O
significance	significance	NN	I-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
by	by	IN	B-PP	O
using	use	VBG	B-VP	O
a	a	DT	B-NP	O
neural	neural	JJ	I-NP	O

network	network	NN	B-NP	O
to	to	TO	B-PP	O
train	train	NN	B-NP	B-protein
2	2	CD	I-NP	I-protein
,	,	,	I-NP	O
182	182	CD	I-NP	O
scoring	score	VBG	I-NP	O
siRNAs	siRNA	NNS	I-NP	O
(	(	(	O	O
scores	score	NNS	B-NP	O
are	be	VBP	B-VP	O
real	real	JJ	B-NP	O
numbers	number	NNS	I-NP	O
in	in	IN	B-PP	O
[	[	(	B-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
the	the	DT	B-NP	O
high	high	JJ	I-NP	O

score	score	VBP	B-VP	O
the	the	DT	B-NP	O
higher	high	JJR	I-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
)	)	)	O	O
and	and	CC	O	O
test	test	NN	B-NP	O
on	on	IN	B-PP	O
249	249	CD	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
[	[	(	O	O
5	5	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
This	This	DT	B-NP	O
data	datum	NNS	I-NP	O
set	set	VBD	B-VP	O

was	be	VBD	B-VP	O
consequently	consequently	RB	I-VP	O
used	use	VBN	I-VP	O
to	to	TO	B-VP	O
build	build	VB	I-VP	O
other	other	JJ	B-NP	O
predictive	predictive	JJ	I-NP	O
models	model	NNS	I-NP	O
[	[	(	O	O
6	6	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
13	13	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
[	[	(	O	O
16	16	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
Recently	Recently	RB	B-ADVP	O
,	,	,	O	O

Qui	Qui	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
used	use	VBN	B-VP	O
multiple	multiple	JJ	B-NP	O
support	support	NN	I-NP	O
vector	vector	NN	I-NP	O
regression	regression	NN	I-NP	O
with	with	IN	B-PP	O
RNA	RNA	NN	B-NP	O
string	string	VBG	B-VP	O
kernel	kernel	NN	B-NP	O
for	for	IN	B-PP	O

siRNA	siRNA	NN	B-NP	O
efficacy	efficacy	NN	I-NP	O
prediction	prediction	NN	I-NP	O
[	[	(	O	O
8	8	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
and	and	CC	O	O
Sciabola	Sciabola	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
applied	apply	VBD	B-VP	O
three	three	CD	B-NP	O
dimension	dimension	NN	I-NP	O
struc-	struc-	NN	I-NP	O

tural	tural	JJ	B-NP	O
information	information	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
to	to	TO	B-VP	O
increase	increase	VB	I-VP	O
predictability	predictability	NN	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
regression	regression	NN	I-NP	O
model	model	NN	I-NP	O
[	[	(	O	O
12	12	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O

However	However	RB	B-ADVP	O
,	,	,	O	O
most	most	JJS	B-NP	O
of	of	IN	B-PP	O
those	those	DT	B-NP	O
methods	method	NNS	I-NP	O
suffer	suffer	NN	I-NP	O
from	from	IN	B-PP	O
some	some	DT	B-NP	O
drawbacks.	drawbacks.	FW	I-NP	O
Their	Their	PRP$	B-NP	O
correlations	correlation	NNS	I-NP	O

between	between	IN	B-PP	O
predicted	predict	VBN	B-NP	O
values	value	NNS	I-NP	O
and	and	CC	O	O
experimental	experimental	JJ	B-NP	O
values	value	NNS	I-NP	O
of	of	IN	B-PP	O
dependent	dependent	JJ	B-NP	O
variable	variable	NN	I-NP	O
ranging	range	VBG	B-VP	O

from	from	IN	B-PP	O
0.60	0.60	CD	B-NP	O
to	to	TO	B-PP	O
0.68	0.68	CD	B-NP	O
were	be	VBD	B-VP	O
considerably	considerably	RB	I-VP	O
decreased	decrease	VBN	I-VP	O
when	when	WRB	B-ADVP	O
testing	test	VBG	B-VP	O
on	on	IN	B-PP	O
independent	independent	JJ	B-NP	O
data	datum	NNS	I-NP	O

sets.	sets.	LS	B-LST	O
It	It	PRP	B-NP	O
may	may	MD	B-VP	O
be	be	VB	I-VP	O
caused	cause	VBN	I-VP	O
by	by	IN	B-PP	O
the	the	DT	B-NP	O
fact	fact	NN	I-NP	O
that	that	IN	B-SBAR	O
the	the	DT	B-NP	O
Huesken	Huesken	NNP	I-NP	O
dataset	dataset	NN	I-NP	O
may	may	MD	B-VP	O
not	not	RB	I-VP	O
be	be	VB	I-VP	O
repre-	repre-	NN	B-NP	O

sentative	sentative	JJ	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	O
population	population	NN	I-NP	O
having	have	VBG	B-VP	O
about	about	IN	B-NP	O
419	419	CD	I-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
and	and	CC	O	O
the	the	DT	B-NP	O
sample	sample	NN	I-NP	O
size	size	NN	I-NP	O

is	be	VBZ	B-VP	O
small.	small.	JJ	B-ADJP	O
Besides	Besides	IN	B-PP	O
the	the	DT	B-NP	O
scoring	score	VBG	I-NP	O
siRNA	siRNA	NN	I-NP	O
dataset	dataset	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
labelled	label	VBN	I-NP	O
siRNA	siRNA	NN	I-NP	O
datasets	dataset	NNS	I-NP	O
,	,	,	O	O
e.g	e.g	NN	B-NP	O
.	.	.	O	O

siRecord	siRecord	NN	B-NP	O
database	database	NN	I-NP	O
[	[	(	O	O
9	9	CD	B-NP	O
]	]	)	O	O
with	with	IN	B-PP	O
labels	label	NNS	B-NP	O
such	such	JJ	B-PP	O
as	as	IN	I-PP	O
‘very	‘very	JJ	B-NP	O
high”	high”	NN	I-NP	O
,	,	,	O	O
‘high’	‘high’	NN	B-NP	O
,	,	,	O	O
‘medium’	‘medium’	NN	B-NP	O
,	,	,	O	O
‘low’	‘low’	NN	B-NP	B-protein
for	for	IN	B-PP	O

the	the	DT	B-NP	O
knockdown	knockdown	JJ	I-NP	O
ability	ability	NN	I-NP	O
were	be	VBD	B-VP	O
also	also	RB	I-VP	O
exploited	exploit	VBN	I-VP	O
by	by	IN	B-PP	O
classification	classification	NN	B-NP	O
methods	method	NNS	I-NP	O
.	.	.	O	O


Our	Our	PRP$	B-NP	O
work	work	NN	I-NP	O
aims	aim	VBZ	B-VP	O
to	to	TO	I-VP	O
develop	develop	VB	I-VP	O
a	a	DT	B-NP	O
novel	novel	JJ	I-NP	O
framework	framework	NN	I-NP	O
for	for	IN	B-PP	O
better	good	JJR	B-NP	O
prediction	prediction	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O

siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	NN	I-NP	O
ability.	ability.	IN	B-PP	O
The	The	DT	B-NP	O
key	key	JJ	I-NP	O
idea	idea	NN	I-NP	O
is	be	VBZ	B-VP	O
not	not	RB	B-VP	O
only	only	RB	I-VP	O
focusing	focus	VBG	I-VP	O
on	on	IN	B-PP	O
learning	learn	VBG	B-VP	O
algo-	algo-	RP	B-PRT	O

rithms	rithm	NNS	B-NP	O
but	but	CC	O	O
also	also	RB	B-ADVP	O
exploiting	exploit	VBG	B-VP	O
results	result	NNS	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
empirical	empirical	JJ	I-NP	O
process	process	NN	I-NP	O
to	to	TO	B-VP	O
enrich	enrich	VB	I-VP	O
the	the	DT	B-NP	O
data.	data.	NN	I-NP	O
To	To	TO	O	O

this	this	DT	B-NP	O
end	end	NN	I-NP	O
,	,	,	O	O
we	we	PRP	B-NP	O
first	first	VBP	B-VP	O
learn	learn	VBP	B-VP	O
transformation	transformation	NN	B-NP	O
matrices	matrix	NNS	I-NP	O
by	by	IN	B-PP	O
incorporating	incorporate	VBG	B-VP	O
existing	exist	VBG	B-NP	O
siRNA	siRNA	NN	I-NP	O

design	design	NN	B-NP	O
rules	rule	NNS	I-NP	O
with	with	IN	B-PP	O
labelled	label	VBN	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
in	in	IN	B-PP	O
siRecord	siRecord	NN	B-NP	O
database.	database.	NN	I-NP	O
We	We	PRP	B-NP	O
then	then	RB	B-ADVP	O
use	use	VBP	B-VP	O
the	the	DT	B-NP	O
trans-	trans-	NN	I-NP	O

formation	formation	NN	B-NP	O
matrices	matrix	NNS	I-NP	O
to	to	TO	B-VP	O
enrich	enrich	VB	I-VP	O
scoring	score	VBG	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
as	as	IN	B-PP	O
transformed	transform	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
and	and	CC	O	O
do	do	VB	B-VP	O

prediction	prediction	NN	B-NP	O
with	with	IN	B-PP	O
them	them	PRP	B-NP	O
by	by	IN	B-PP	O
bilinear	bilinear	JJ	B-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O
where	where	WRB	B-ADVP	O
the	the	DT	B-NP	O
Frobenius	Frobenius	NNP	I-NP	O
norm	norm	NN	I-NP	O
is	be	VBZ	B-VP	O

appropriately	appropriately	RB	B-ADVP	O
replaced	replace	VBN	B-VP	O
by	by	IN	B-PP	O
L2	L2	NN	B-NP	O
regularization	regularization	NN	I-NP	O
norm	norm	NN	I-NP	O
for	for	IN	B-PP	O
an	an	DT	B-NP	O
effective	effective	JJ	I-NP	O
computation	computation	NN	I-NP	O
.	.	.	O	O

Experiments	Experiment	NNS	B-NP	O
show	show	VBP	B-VP	O
that	that	IN	B-SBAR	O
the	the	DT	B-NP	O
proposed	propose	VBN	I-NP	O
method	method	NN	I-NP	O
achieves	achieve	VBZ	B-VP	O
results	result	NNS	B-NP	O
better	good	JJR	B-ADJP	O
than	than	IN	B-PP	O
most	most	JJS	B-NP	O

existing	exist	VBG	B-NP	O
models.	models.	NNS	I-NP	O
The	The	DT	B-NP	O
contributions	contribution	NNS	I-NP	O
of	of	IN	B-PP	O
this	this	DT	B-NP	O
work	work	NN	I-NP	O
are	be	VBP	B-VP	O
summarized	summarize	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O


1.	1.	LS	B-LST	O
A	A	DT	B-NP	O
novel	novel	JJ	I-NP	O
generic	generic	JJ	I-NP	O
framework	framework	NN	I-NP	O
to	to	TO	B-VP	O
predict	predict	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
efficacy	efficacy	NN	I-NP	O
by	by	IN	B-PP	O
enriching	enrich	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	O

sequences	sequence	NNS	B-NP	O
with	with	IN	B-PP	O
domain	domain	NN	B-NP	O
knowledge	knowledge	NN	I-NP	O
and	and	CC	O	O
appropriately	appropriately	RB	B-VP	O
using	use	VBG	I-VP	O
bilinear	bilinear	JJ	B-NP	O
tensor	tensor	NN	I-NP	O

regression	regression	NN	B-NP	O
.	.	.	O	O


2.	2.	LS	B-LST	O
An	An	DT	B-NP	O
optimization	optimization	NN	I-NP	O
method	method	NN	I-NP	O
to	to	TO	B-VP	O
enrich	enrich	VB	I-VP	O
siRNAs	siRNA	NNS	B-NP	B-RNA
using	use	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
found	find	VBD	B-VP	O


by	by	IN	B-PP	O
empirical	empirical	JJ	B-NP	O
works	work	NNS	I-NP	O
.	.	.	O	O


3.	3.	LS	B-LST	O
The	The	DT	B-NP	O
use	use	NN	I-NP	O
of	of	IN	B-PP	O
L2	L2	NN	B-NP	B-DNA
norm	norm	NN	I-NP	O
instead	instead	RB	B-PP	O
of	of	IN	I-PP	O
Frobenius	Frobenius	NNP	B-NP	O
norm	norm	NN	I-NP	O
in	in	IN	B-PP	O
bilinear	bilinear	JJ	B-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O


that	that	DT	B-NP	O
allows	allow	VBZ	B-VP	O
effectively	effectively	RB	I-VP	O
learning	learn	VBG	I-VP	O
the	the	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
model	model	NN	B-NP	O
parameters	parameter	NNS	I-NP	O
.	.	.	O	O


402	402	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


2	2	LS	B-LST	O
The	The	DT	B-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


The	The	DT	B-NP	O
problem	problem	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
prediction	prediction	NN	I-NP	O
using	use	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O

is	be	VBZ	B-VP	O
formulated	formulate	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O
:	:	:	O	O


–	–	RB	B-ADVP	O
Given	Give	VBN	B-VP	O
:	:	:	O	O
Two	Two	CD	B-NP	O
sets	set	NNS	I-NP	O
of	of	IN	B-PP	O
labelled	label	VBN	B-NP	O
siRNA	siRNA	NN	I-NP	O
and	and	CC	O	O
scoring	score	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
of	of	IN	B-PP	O
length	length	NN	B-NP	O
n	n	NN	I-NP	O
,	,	,	O	O


and	and	CC	O	O
a	a	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
K	K	NN	B-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
.	.	.	O	O


–	–	IN	B-PP	O
Find	Find	NNP	B-NP	O
:	:	:	O	O
A	A	DT	B-NP	O
function	function	NN	I-NP	O
that	that	WDT	B-NP	O
assigns	assign	VBZ	B-VP	O
a	a	DT	B-NP	O
right	right	JJ	I-NP	O
score	score	NN	I-NP	O
to	to	TO	B-PP	O
a	a	DT	B-NP	O
given	give	VBN	I-NP	O
siRNA	siRNA	NN	I-NP	O
.	.	.	O	O


The	The	DT	B-NP	O
proposed	propose	VBN	I-NP	O
framework	framework	NN	I-NP	O
consists	consist	VBZ	B-VP	O
of	of	IN	B-PP	O
four	four	CD	B-NP	O
steps	step	NNS	I-NP	O
in	in	IN	B-PP	O
two	two	CD	B-NP	O
phases.	phases.	NNS	I-NP	O
The	The	DT	B-NP	O
first	first	NN	I-NP	O
phase	phase	NN	I-NP	O

is	be	VBZ	B-VP	O
to	to	TO	B-VP	O
encode	encode	VB	I-VP	O
siRNAs	siRNA	NNS	B-NP	O
and	and	CC	O	O
learn	learn	VB	B-VP	O
transformation	transformation	NN	B-NP	O
matrices.	matrices.	NN	I-NP	O
The	The	DT	B-NP	O
second	second	JJ	I-NP	O
phase	phase	NN	I-NP	O
is	be	VBZ	B-VP	O

to	to	TO	B-VP	O
use	use	VB	I-VP	O
transformation	transformation	NN	B-NP	O
matrices	matrix	NNS	I-NP	O
to	to	TO	B-VP	O
enrich	enrich	VB	I-VP	O
siRNAs	siRNA	NNS	B-NP	O
as	as	IN	B-PP	O
transformed	transform	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
and	and	CC	O	O

learn	learn	NN	B-NP	O
model	model	NN	I-NP	O
parameters	parameter	NNS	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
bilinear	bilinear	JJ	I-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O
to	to	TO	B-VP	O
predict	predict	VB	I-VP	O
the	the	DT	B-NP	O
score	score	NN	I-NP	O
of	of	IN	B-PP	O

siRNAs	siRNA	NNS	B-NP	O
using	use	VBG	B-VP	O
transformed	transform	VBN	B-NP	O
matrices.	matrices.	NN	I-NP	O
The	The	DT	B-NP	O
steps	step	NNS	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
framework	framework	NN	I-NP	O
are	be	VBP	B-VP	O
summarized	summarize	VBN	I-VP	O

in	in	IN	B-PP	O
Table	Table	JJ	B-NP	B-protein
1	1	CD	I-NP	I-protein
.	.	.	O	O


Table	Table	JJ	B-NP	O
1.	1.	CD	I-NP	O
Framework	Framework	NN	I-NP	O
for	for	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
prediction	prediction	NN	I-NP	O


1.	1.	LS	B-LST	O
To	To	TO	B-VP	O
encode	encode	VB	I-VP	O
each	each	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA
as	as	IN	B-PP	O
an	an	DT	B-NP	O
encoding	encode	VBG	I-NP	O
matrix	matrix	NN	I-NP	B-protein
X	X	NN	I-NP	I-protein
representing	represent	VBG	B-VP	O
the	the	DT	B-NP	O
nu-	nu-	NN	I-NP	O

cleotides	cleotide	NNS	B-NP	O
A	A	NN	I-NP	O
,	,	,	O	O
C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O
U	U	NN	B-NP	O
at	at	IN	B-PP	O
n	n	NN	B-NP	O
positions	position	NNS	I-NP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
sequence.	sequence.	NN	I-NP	O
Thus	Thus	RB	B-ADVP	O
,	,	,	O	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA

are	be	VBP	B-VP	O
represented	represent	VBN	I-VP	O
as	as	IN	B-PP	O
n	n	NN	B-NP	B-protein
×	×	NN	I-NP	I-protein
4	4	CD	I-NP	I-protein
encoding	encode	VBG	I-NP	O
matrices	matrix	NNS	I-NP	O
.	.	.	O	O


2.	2.	LS	B-LST	O
To	To	TO	B-VP	O
learn	learn	VB	I-VP	O
transformation	transformation	NN	B-NP	B-protein
matrices	matrix	NNS	I-NP	I-protein
Tk	Tk	NN	I-NP	I-protein
,	,	,	O	O
k	k	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
K	K	NN	B-NP	O
,	,	,	O	O
each	each	DT	B-NP	O
characterizes	characterize	VBZ	B-VP	O
the	the	DT	B-NP	O
knock-	knock-	NN	I-NP	O

down	down	RB	B-ADVP	O
ability	ability	NN	B-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
A	A	NN	I-NP	O
,	,	,	O	O
C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O
U	U	NN	B-NP	O
at	at	IN	B-PP	O
n	n	NN	B-NP	O
positions	position	NNS	I-NP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA

regarding	regard	VBG	B-PP	O
the	the	DT	B-NP	O
kth	kth	NN	I-NP	O
design	design	NN	I-NP	O
rule.	rule.	NN	I-NP	O
Each	Each	DT	B-NP	O
Tk	Tk	NN	I-NP	B-protein
is	be	VBZ	B-VP	O
learned	learn	VBN	I-VP	O
from	from	IN	B-PP	O
the	the	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
labelled	label	VBN	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA

and	and	CC	O	O
the	the	DT	B-NP	O
kth	kth	NN	I-NP	O
design	design	NN	I-NP	O
rule.	rule.	NN	I-NP	O
This	This	DT	B-NP	O
incorporation	incorporation	NN	I-NP	O
of	of	IN	B-PP	O
each	each	DT	B-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
with	with	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	O
leads	lead	VBZ	B-VP	O

to	to	TO	B-VP	O
solve	solve	VB	I-VP	O
a	a	DT	B-NP	O
newly	newly	RB	I-NP	O
formulated	formulate	VBN	I-NP	O
optimization	optimization	NN	I-NP	O
problem	problem	NN	I-NP	O
.	.	.	O	O

3.	3.	LS	B-LST	O
To	To	TO	B-VP	O
transform	transform	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
(	(	(	O	O
encoding	encode	VBG	B-VP	O
matrices	matrix	NNS	B-NP	O
)	)	)	O	O
to	to	TO	B-PP	O
transformed	transform	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
by	by	IN	B-PP	O
K	K	NN	B-NP	O
transfor-	transfor-	NN	I-NP	O

mation	mation	NN	B-NP	O
matrices.	matrices.	NN	I-NP	O
The	The	DT	B-NP	O
transformed	transform	VBN	I-NP	O
matrices	matrix	NNS	I-NP	O
of	of	IN	B-PP	O
size	size	NN	B-NP	O
K	K	NN	I-NP	O
×	×	NN	I-NP	O
n	n	NN	I-NP	O
are	be	VBP	B-VP	O
considered	consider	VBN	I-VP	O
as	as	IN	B-PP	O
second	second	JJ	B-NP	O

order	order	NN	B-NP	O
tensor	tensor	NN	I-NP	O
representations	representation	NNS	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
.	.	.	O	O


4.	4.	LS	B-LST	O
To	To	TO	B-VP	O
build	build	VB	I-VP	O
a	a	DT	B-NP	O
bilinear	bilinear	JJ	I-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O
model	model	NN	I-NP	O
that	that	WDT	B-NP	O
uses	use	VBZ	B-VP	O
transformed	transform	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
of	of	IN	B-PP	O
scor-	scor-	NN	B-NP	O


ing	ing	VBG	B-VP	O
siRNAs	siRNA	NNS	B-NP	O
to	to	TO	B-VP	O
predict	predict	VB	I-VP	O
the	the	DT	B-NP	O
knockdown	knockdown	JJ	I-NP	O
ability	ability	NN	I-NP	O
of	of	IN	B-PP	O
new	new	JJ	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
.	.	.	O	O


2.1	2.1	CD	B-NP	O
Encoding	Encode	VBG	I-NP	O
siRNA	siRNA	NN	I-NP	O
and	and	CC	I-NP	O
Transformation	Transformation	NN	I-NP	O
Matrix	Matrix	NNP	I-NP	O
Learning	Learning	NNP	I-NP	O


Step	Step	NN	B-NP	B-protein
1	1	CD	I-NP	I-protein
of	of	IN	B-PP	O
the	the	DT	B-NP	O
framework	framework	NN	I-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
easily	easily	RB	I-VP	O
done	do	VBN	I-VP	O
where	where	WRB	B-ADVP	O
each	each	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA
with	with	IN	B-PP	O
n	n	NN	B-NP	O

nucleotides	nucleotide	NNS	B-NP	O
in	in	IN	B-PP	O
length	length	NN	B-NP	O
is	be	VBZ	B-VP	O
encoded	encode	VBN	I-VP	O
as	as	IN	B-PP	O
a	a	DT	B-NP	O
binary	binary	JJ	I-NP	O
encoding	encode	VBG	I-NP	O
matrix	matrix	NN	I-NP	O
of	of	IN	B-PP	O
size	size	NN	B-NP	O
n	n	NN	I-NP	O
×	×	IN	B-PP	O
4.	4.	CD	B-NP	O
In	In	IN	O	O

fact	fact	NN	B-NP	O
,	,	,	O	O
four	four	CD	B-NP	O
nucleotides	nucleotide	NNS	I-NP	O
A	A	NN	I-NP	O
,	,	,	O	O
C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
or	or	CC	O	O
U	U	NN	B-NP	O
are	be	VBP	B-VP	O
encoded	encode	VBN	I-VP	O
by	by	IN	B-PP	O
encoding	encode	VBG	B-VP	O
vectors	vector	NNS	B-NP	O
(	(	(	O	O
1	1	CD	B-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
)	)	)	O	O
,	,	,	O	O

(	(	(	O	O
0	0	CD	B-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
)	)	)	O	O
,	,	,	O	O
(	(	(	O	O
0	0	CD	B-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
)	)	)	O	O
and	and	CC	O	O
(	(	(	O	O
0	0	CD	B-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
0	0	CD	I-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
)	)	)	O	O
,	,	,	O	O
respectively.	respectively.	VBP	B-VP	O
If	If	IN	B-SBAR	O
a	a	DT	B-NP	O
nucleotide	nucleotide	NN	I-NP	O
from	from	IN	B-PP	O
A	A	NN	B-NP	O
,	,	,	O	O
C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O

U	U	NN	B-NP	O
appears	appear	VBZ	B-VP	O
at	at	IN	B-PP	O
the	the	DT	B-NP	O
jth	jth	JJ	I-NP	O
position	position	NN	I-NP	O
in	in	IN	B-PP	O
a	a	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA
,	,	,	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
n	n	NN	B-NP	O
,	,	,	O	O
its	its	PRP$	B-NP	O
encoding	encode	VBG	B-VP	O

vector	vector	NN	B-NP	O
will	will	MD	B-VP	O
be	be	VB	I-VP	O
used	use	VBN	I-VP	O
to	to	TO	B-VP	O
encode	encode	VB	I-VP	O
the	the	DT	B-NP	O
jth	jth	JJ	I-NP	O
row	row	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
encoding	encode	VBG	I-NP	O
matrix	matrix	NN	I-NP	O
.	.	.	O	O


Step	Step	NN	B-NP	B-protein
2	2	CD	I-NP	I-protein
is	be	VBZ	B-VP	O
to	to	TO	B-VP	O
learn	learn	VB	I-VP	O
transformation	transformation	NN	B-NP	O
matrices	matrix	NNS	I-NP	O
Tk	Tk	VBP	B-VP	O
regarding	regard	VBG	I-VP	O
the	the	DT	B-NP	O
kth	kth	NN	I-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O

k	k	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
Tk	Tk	NNP	I-NP	O
has	have	VBZ	B-VP	O
size	size	NN	B-NP	O
of	of	IN	B-PP	O
4	4	CD	B-NP	O
×	×	NN	I-NP	O
n	n	NN	I-NP	O
where	where	WRB	B-ADVP	O
the	the	DT	B-NP	O
rows	row	NNS	I-NP	O
correspond	correspond	VBP	B-VP	O
to	to	TO	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
A	A	NN	I-NP	O
,	,	,	O	O

C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O
U	U	NN	B-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
columns	column	NNS	I-NP	O
correspond	correspond	VBP	B-VP	O
to	to	TO	I-VP	O
n	n	VB	I-VP	O
positions	position	NNS	B-NP	O
on	on	IN	B-PP	O
sequences.	sequences.	NN	B-NP	O
Tk	Tk	NN	I-NP	O
are	be	VBP	B-VP	O

learned	learn	VBD	B-VP	O
one	one	CD	B-NP	O
by	by	IN	B-PP	O
one	one	CD	B-NP	O
from	from	IN	B-PP	O
the	the	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	B-RNA
and	and	CC	O	O
the	the	DT	B-NP	O
kth	kth	NN	I-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
thus	thus	RB	O	O
we	we	PRP	B-NP	O

use	use	NN	B-NP	O
T	T	NN	I-NP	O
instead	instead	RB	B-CONJP	O
of	of	IN	I-CONJP	O
Tk	Tk	NN	B-NP	B-protein
for	for	IN	B-PP	O
simplification.	simplification.	NN	B-NP	O
Each	Each	DT	B-NP	O
cell	cell	NN	I-NP	O
T	T	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
,	,	,	O	O
i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
4	4	CD	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
n	n	NN	B-NP	O
,	,	,	O	O

represents	represent	VBZ	B-VP	O
the	the	DT	B-NP	O
knockdown	knockdown	JJ	I-NP	O
ability	ability	NN	I-NP	O
of	of	IN	B-PP	O
nucleotide	nucleotide	NN	B-NP	O
i	i	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	O
j	j	NN	I-NP	O
regarding	regard	VBG	B-VP	O
the	the	DT	B-NP	O
kth	kth	NN	I-NP	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


403	403	CD	B-NP	O


Fig.	Fig.	NNP	B-NP	O
1.	1.	LS	B-LST	O
The	The	DT	B-NP	O
left	left	JJ	I-NP	O
table	table	NN	I-NP	O
shows	show	VBZ	B-VP	O
an	an	DT	B-NP	O
example	example	NN	I-NP	O
of	of	IN	B-PP	O
encoding	encode	VBG	B-VP	O
matrix	matrix	NN	B-NP	O
,	,	,	O	O
transformation	transformation	NN	B-NP	B-protein
matrix	matrix	NN	I-NP	I-protein
,	,	,	O	O
and	and	CC	O	O

transformed	transform	VBN	B-NP	O
vector	vector	NN	I-NP	O
(	(	(	O	O
the	the	DT	B-NP	O
values	value	NNS	I-NP	O
0.5	0.5	CD	B-NP	O
,	,	,	O	O
0.1	0.1	CD	B-NP	O
etc.	etc.	FW	I-NP	O
are	be	VBP	B-VP	O
taken	take	VBN	I-VP	O
to	to	TO	B-PP	O
the	the	DT	B-NP	O
transformed	transform	VBN	I-NP	O
vector	vector	NN	I-NP	O
)	)	)	O	O
.	.	.	O	O
The	The	DT	O	O

right	right	JJ	B-NP	O
table	table	NN	I-NP	O
is	be	VBZ	B-VP	O
an	an	DT	B-NP	O
example	example	NN	I-NP	O
of	of	IN	B-PP	O
incorporating	incorporate	VBG	B-VP	O
the	the	DT	B-NP	O
condition	condition	NN	I-NP	O
of	of	IN	B-PP	O
a	a	DT	B-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
19	19	CD	I-NP	I-protein

to	to	TO	B-PP	O
a	a	DT	B-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
T	T	NN	I-NP	O
by	by	IN	B-PP	O
designing	design	VBG	B-VP	O
constraints	constraint	NNS	B-NP	O
.	.	.	O	O


design	design	NN	B-NP	O
rule.	rule.	NN	I-NP	O
Each	Each	DT	B-NP	O
cell	cell	NN	I-NP	O
T	T	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
to	to	TO	B-VP	O
be	be	VB	I-VP	O
learned	learn	VBN	I-VP	O
have	have	VBP	B-VP	O
to	to	TO	I-VP	O
satisfy	satisfy	VB	I-VP	O
a	a	DT	B-NP	O
number	number	NN	I-NP	O
of	of	IN	B-PP	O
constraints	constraint	NNS	B-NP	O
.	.	.	O	O

First	First	RB	B-ADVP	O
,	,	,	O	O
they	they	PRP	B-NP	O
are	be	VBP	B-VP	O
basic	basic	JJ	B-NP	O
and	and	CC	I-NP	O
normalization	normalization	NN	I-NP	O
constraints	constraint	NNS	I-NP	O
on	on	IN	B-PP	O
elements	element	NNS	B-NP	O
of	of	IN	B-PP	O
T	T	NN	B-NP	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
≥	≥	NN	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O

i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
4	4	CD	B-NP	O
;	;	:	O	O

i=1	i=1	NN	B-NP	O
T	T	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
n	n	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O


j	j	NN	B-NP	O
=	=	SYM	B-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
2	2	CD	I-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
n	n	NN	B-NP	O


(	(	(	O	O
1	1	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
2	2	CD	B-NP	O
)	)	)	O	O


The	The	DT	B-NP	O
second	second	JJ	I-NP	O
kind	kind	NN	I-NP	O
of	of	IN	B-PP	O
constraints	constraint	NNS	B-NP	O
related	relate	VBN	B-VP	O
to	to	TO	I-VP	O
design	design	VB	I-VP	O
rules.	rules.	NN	B-NP	B-protein
Each	Each	DT	B-NP	I-protein
design	design	NN	I-NP	I-protein
rule	rule	NN	I-NP	I-protein
propo-	propo-	NN	I-NP	I-protein

sitionally	sitionally	RB	B-ADVP	O
describes	describe	VBZ	B-VP	O
the	the	DT	B-NP	O
occurrence	occurrence	NN	I-NP	O
or	or	CC	I-NP	O
absence	absence	NN	I-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
at	at	IN	B-PP	O
different	different	JJ	B-NP	O
positions	position	NNS	I-NP	O

of	of	IN	B-PP	O
effective	effective	JJ	B-NP	O
siRNA	siRNA	NN	I-NP	O
sequences.	sequences.	NN	I-NP	O
Therefore	Therefore	RB	B-ADVP	O
,	,	,	O	O
if	if	IN	B-SBAR	O
a	a	DT	B-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
shows	show	VBZ	B-VP	O
the	the	DT	B-NP	O
occurrence	occurrence	NN	I-NP	O

(	(	(	O	O
absence	absence	NN	B-NP	O
)	)	)	O	O
of	of	IN	B-PP	O
some	some	DT	B-NP	O
nucleotides	nucleotide	NNS	I-NP	O
on	on	IN	B-PP	O
jth	jth	NN	B-NP	O
position	position	NN	I-NP	O
,	,	,	O	O
then	then	RB	O	O
their	their	PRP$	B-NP	O
corresponding	correspond	VBG	I-NP	O
values	value	NNS	I-NP	O

in	in	IN	B-PP	O
the	the	DT	B-NP	O
matrix	matrix	NN	I-NP	O
T	T	NN	I-NP	O
would	would	MD	B-VP	O
be	be	VB	I-VP	O
greater	great	JJR	B-ADJP	O
(	(	(	O	O
smaller	small	JJR	B-ADJP	O
)	)	)	O	O
than	than	IN	B-PP	O
other	other	JJ	B-NP	O
values	value	NNS	I-NP	O
at	at	IN	B-PP	O
column	column	NN	B-NP	O
j.	j.	NN	I-NP	O
For	For	IN	O	O

example	example	NN	B-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
right	right	JJ	I-NP	O
table	table	NN	I-NP	O
in	in	IN	B-PP	O
Figure	Figure	NN	B-NP	O
1	1	CD	I-NP	O
illustrates	illustrate	VBZ	B-VP	O
that	that	IN	B-SBAR	O
at	at	IN	B-PP	O
posi-	posi-	NN	B-NP	O

tion	tion	NN	B-NP	O
19	19	CD	I-NP	O
,	,	,	O	O
nucleotides	nucleotide	NNS	B-NP	O
A/U	A/U	NN	I-NP	O
are	be	VBP	B-VP	O
effective	effective	JJ	B-NP	O
and	and	CC	I-NP	O
nucleotide	nucleotide	NN	I-NP	O
C	C	NN	I-NP	O
is	be	VBZ	B-VP	O
ineffective.	ineffective.	FW	B-ADJP	O
It	It	PRP	B-NP	O
means	mean	VBZ	B-VP	O

that	that	IN	B-SBAR	O
knockdown	knockdown	VBN	B-NP	O
ability	ability	NN	I-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
A/U	A/U	NN	I-NP	O
are	be	VBP	B-VP	O
bigger	big	JJR	B-ADJP	O
than	than	IN	B-PP	O
that	that	DT	B-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O

G/C	G/C	NN	B-NP	O
and	and	CC	I-NP	O
knockdown	knockdown	JJ	I-NP	O
ability	ability	NN	I-NP	O
of	of	IN	B-PP	O
nucleotide	nucleotide	NN	B-NP	O
C	C	NN	I-NP	O
is	be	VBZ	B-VP	O
smaller	small	JJR	B-ADJP	O
than	than	IN	B-PP	O
that	that	DT	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
other	other	JJ	I-NP	O

nucleotides.	nucleotides.	RB	B-ADVP	O
Thus	Thus	RB	I-ADVP	O
,	,	,	O	O
values	value	NNS	B-NP	O
T	T	NN	I-NP	O
[	[	(	I-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
19	19	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	O	O
2	2	CD	B-NP	O
,	,	,	I-NP	O
19	19	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	B-NP	O
3	3	CD	I-NP	O
,	,	,	I-NP	O
19	19	CD	I-NP	O
]	]	)	O	O
and	and	CC	O	O
T	T	NN	B-NP	O
[	[	(	B-NP	O
4	4	CD	I-NP	O
,	,	,	I-NP	O
19	19	CD	I-NP	O
]	]	)	O	O
show	show	VBP	B-VP	O
the	the	DT	B-NP	O
knock-	knock-	NN	I-NP	O

down	down	RB	B-ADVP	O
ability	ability	NN	B-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
A	A	NN	I-NP	O
,	,	,	O	O
C	C	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
and	and	CC	I-NP	O
U	U	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
19	19	CD	I-NP	I-protein
,	,	,	O	O
respectively.	respectively.	RB	B-ADVP	O
Therefore	Therefore	RB	I-ADVP	O
,	,	,	O	O

five	five	VB	B-VP	O
constraints	constraint	NNS	B-NP	O
at	at	IN	B-PP	O
column	column	NN	B-NP	O
19	19	CD	I-NP	O
of	of	IN	B-PP	O
T	T	NN	B-NP	O
are	be	VBP	B-VP	O
formed.	formed.	VBN	I-VP	O
Generally	Generally	RB	B-ADVP	O
,	,	,	O	O
we	we	PRP	B-NP	O
denote	denote	VBP	B-VP	O
the	the	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
R	R	NN	B-NP	B-protein

trick	trick	NN	B-NP	O
inequality	inequality	NN	I-NP	O
constraints	constraint	NNS	I-NP	O
on	on	IN	B-PP	O
T	T	NN	B-NP	O
by	by	IN	B-PP	O
the	the	DT	B-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
under	under	IN	B-PP	O
consideration	consideration	NN	B-NP	O
by	by	IN	B-PP	O


{	{	NN	B-NP	B-protein
gr	gr	NN	I-NP	I-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O
<	<	SYM	O	O
0	0	CD	B-NP	O
}	}	NN	I-NP	O
R	R	NN	I-NP	O


r=1	r=1	NN	B-NP	O


(	(	(	O	O
3	3	CD	B-NP	O
)	)	)	O	O


The	The	DT	B-NP	O
third	third	JJ	I-NP	O
kind	kind	NN	I-NP	O
of	of	IN	B-PP	O
constraints	constraint	NNS	B-NP	O
relating	relate	VBG	B-VP	O
to	to	TO	B-PP	O
preservation	preservation	NN	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	O
classes	class	NNS	I-NP	O
after	after	IN	O	O

being	be	VBG	B-VP	O
transformed	transform	VBN	I-VP	O
by	by	IN	B-PP	O
using	use	VBG	B-VP	O
transformation	transformation	NN	B-NP	B-protein
matrices	matrix	NNS	I-NP	I-protein
Tk	Tk	NN	I-NP	I-protein
,	,	,	O	O
it	it	PRP	B-NP	O
means	mean	VBZ	B-VP	O
that	that	IN	B-SBAR	O
siRNAs	siRNA	NNS	B-NP	O

belonging	belong	VBG	B-VP	O
to	to	TO	B-PP	O
the	the	DT	B-NP	O
same	same	JJ	I-NP	O
class	class	NN	I-NP	O
should	should	MD	B-VP	O
be	be	VB	I-VP	O
more	more	RBR	B-ADJP	O
similar	similar	JJ	I-ADJP	O
to	to	TO	B-PP	O
each	each	DT	B-NP	O
other	other	JJ	I-NP	O
than	than	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	O

belonging	belong	VBG	B-VP	O
to	to	TO	B-PP	O
the	the	DT	B-NP	O
other	other	JJ	I-NP	O
class	class	NN	I-NP	O
.	.	.	O	O


Let	Let	NN	B-NP	O
vector	vector	NN	I-NP	O
xl	xl	NN	I-NP	O
of	of	IN	B-PP	O
size	size	NN	B-NP	O
1	1	CD	I-NP	O
×	×	NN	I-NP	O
n	n	NN	I-NP	O
denote	denote	VBP	B-VP	O
the	the	DT	B-NP	O
transformed	transform	VBN	I-NP	O
vector	vector	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
lth	lth	NN	I-NP	O
siRNA	siRNA	NN	I-NP	O

sequence	sequence	NN	B-NP	O
using	use	VBG	B-VP	O
the	the	DT	B-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
T	T	NN	I-NP	O
.	.	VBD	B-VP	O
The	The	DT	B-NP	O
jth	jth	JJ	I-NP	B-DNA
element	element	NN	I-NP	I-DNA
of	of	IN	B-PP	O
xl	xl	NN	B-NP	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
element	element	NN	I-NP	O

of	of	IN	B-PP	O
T	T	NN	B-NP	O
at	at	IN	B-PP	O
column	column	NN	B-NP	O
j	j	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
row	row	NN	I-NP	O
corresponds	correspond	VBZ	B-VP	O
to	to	TO	B-PP	O
the	the	DT	B-NP	O
jth	jth	JJ	I-NP	O
nucleotide	nucleotide	NN	I-NP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	O

sequence.	sequence.	LS	B-LST	O
To	To	TO	B-VP	O
compute	compute	VB	I-VP	O
xl	xl	NN	B-NP	O
,	,	,	O	O
new	new	JJ	B-NP	O
column-wise	column-wise	JJ	I-NP	O
inner	inner	JJ	I-NP	O
product	product	NN	I-NP	O
is	be	VBZ	B-VP	O
defined	define	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O


xl	xl	NN	B-NP	O
=	=	SYM	B-VP	O
T	T	NN	B-NP	O
◦	◦	SYM	B-VP	O
Xl	Xl	NN	B-NP	O
=	=	SYM	O	O
(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xl	Xl	NN	B-NP	O
[	[	(	O	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	O	O
.	.	NN	B-NP	O
,	,	,	O	O
1	1	CD	B-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xl	Xl	NN	B-NP	O
[	[	(	O	O
2	2	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	O	O
.	.	NN	B-NP	O
,	,	,	O	O
2	2	CD	B-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xl	Xl	NN	B-NP	O
[	[	(	O	O
n	n	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
n	n	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
4	4	CD	B-NP	O
)	)	)	O	O


404	404	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


where	where	WRB	B-ADVP	O
Xl	Xl	NN	B-NP	B-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
and	and	CC	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
are	be	VBP	B-VP	O
the	the	DT	B-NP	O
jth	jth	JJ	I-NP	O
row	row	NN	I-NP	O
vector	vector	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
jth	jth	JJ	I-NP	O
column	column	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
matrix	matrix	NN	I-NP	O

Xl	Xl	NN	B-NP	O
and	and	CC	I-NP	O
T	T	NN	I-NP	O
,	,	,	O	O
respectively	respectively	RB	B-ADVP	O
,	,	,	O	O
and	and	CC	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O
x	x	NN	B-NP	O
,	,	,	O	O
y	y	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
denotes	denote	VBZ	B-VP	O
the	the	DT	B-NP	O
inner	inner	JJ	I-NP	O
product	product	NN	I-NP	O
of	of	IN	B-PP	O
vectors	vector	NNS	B-NP	O
x	x	NN	I-NP	O
and	and	CC	I-NP	O
y	y	NN	I-NP	O
.	.	.	O	O


The	The	DT	B-NP	O
left	left	JJ	I-NP	O
table	table	NN	I-NP	O
in	in	IN	B-PP	O
Figure	Figure	NN	B-NP	O
1	1	CD	I-NP	O
shows	show	VBZ	B-VP	O
an	an	DT	B-NP	O
example	example	NN	I-NP	O
of	of	IN	B-PP	O
encoding	encode	VBG	B-VP	O
matrix	matrix	NN	B-NP	B-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
trans-	trans-	NN	B-NP	O

formation	formation	NN	B-NP	O
matrix	matrix	NN	I-NP	O
T	T	NN	I-NP	O
and	and	CC	O	O
transformed	transform	VBN	B-NP	O
vector	vector	NN	I-NP	O
x	x	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
given	give	VBN	I-NP	O
sequence	sequence	NN	I-NP	O
AUGCU	AUGCU	NN	I-NP	O
.	.	.	O	O

The	The	DT	B-NP	O
rows	row	NNS	I-NP	O
of	of	IN	B-PP	O
X	X	NN	B-NP	O
represent	represent	VBP	B-VP	O
encoding	encode	VBG	B-NP	O
vectors	vector	NNS	I-NP	O
of	of	IN	B-PP	O
nucleotides	nucleotide	NNS	B-NP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
sequence.	sequence.	NN	I-NP	O
Given	Give	VBN	B-VP	O

transformation	transformation	NN	B-NP	O
matrix	matrix	NN	I-NP	O
T	T	NN	I-NP	O
of	of	IN	B-PP	O
size	size	NN	B-NP	O
4	4	CD	I-NP	O
×	×	NN	I-NP	O
5.	5.	CD	I-NP	O
The	The	DT	B-NP	O
sequence	sequence	NN	I-NP	O
AUGCU	AUGCU	NN	I-NP	O
is	be	VBZ	B-VP	O
represented	represent	VBN	I-VP	O
by	by	IN	B-PP	O

the	the	DT	B-NP	O
vector	vector	NN	I-NP	O
x	x	NN	I-NP	O
=	=	SYM	B-VP	O
(	(	(	O	O
T	T	NN	B-NP	O
[	[	(	B-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	B-NP	O
4	4	CD	I-NP	O
,	,	,	I-NP	O
1	1	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
3	3	CD	I-NP	O
,	,	,	I-NP	O
3	3	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	O	O
2	2	CD	B-NP	O
,	,	,	I-NP	O
4	4	CD	I-NP	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	O	O
4	4	CD	B-NP	O
,	,	,	O	O
5	5	CD	B-NP	O
]	]	)	O	O
)	)	)	O	O
=	=	JJ	B-ADJP	O
(	(	(	O	O
0.5	0.5	CD	B-NP	O
,	,	,	I-NP	O
0.1	0.1	CD	I-NP	O
,	,	,	I-NP	O
0.08	0.08	CD	I-NP	O
,	,	,	I-NP	O
0.6	0.6	CD	I-NP	O
,	,	,	I-NP	O
0.1	0.1	CD	I-NP	O
)	)	)	O	O
.	.	.	O	O

Therefore	Therefore	RB	B-ADVP	O
,	,	,	O	O
transformed	transform	VBN	B-NP	O
data	datum	NNS	I-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
computed	compute	VBN	I-VP	O
by	by	IN	B-PP	O
the	the	DT	B-NP	O
column-wise	column-wise	JJ	I-NP	B-protein
inner	inner	JJ	I-NP	I-protein
product	product	NN	I-NP	I-protein

x	x	NN	B-NP	O
=	=	JJ	I-NP	O
T	T	NN	I-NP	O
◦	◦	NN	I-NP	O
X	X	NN	I-NP	O
.	.	.	O	O


The	The	DT	B-NP	O
problem	problem	NN	I-NP	O
of	of	IN	B-PP	O
transformation	transformation	NN	B-NP	O
matrix	matrix	NN	I-NP	O
learning	learning	NN	I-NP	O
is	be	VBZ	B-VP	O
now	now	RB	I-VP	O
formulated	formulate	VBN	I-VP	O
as	as	IN	B-PP	O
finding	finde	VBG	B-VP	O
T	T	NN	B-NP	O

under	under	IN	B-PP	O
constraints	constraint	NNS	B-NP	O
(	(	(	O	O
1	1	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
(	(	(	O	O
2	2	CD	B-NP	O
)	)	)	O	O
and	and	CC	O	O
(	(	(	B-LST	O
3	3	LS	I-LST	O
)	)	)	O	O
so	so	IN	B-SBAR	O
that	that	IN	I-SBAR	O
the	the	DT	B-NP	O
similarity	similarity	NN	I-NP	O
of	of	IN	B-PP	O
transformed	transform	VBN	B-NP	O
vectors	vector	NNS	I-NP	O
xl	xl	VBP	B-VP	O

in	in	IN	B-PP	O
the	the	DT	B-NP	O
same	same	JJ	I-NP	O
class	class	NN	I-NP	O
is	be	VBZ	B-VP	O
minimum	minimum	NN	B-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
dissimilarity	dissimilarity	NN	I-NP	O
of	of	IN	B-PP	O
xl	xl	NN	B-NP	O
in	in	IN	B-PP	O
different	different	JJ	B-NP	O
classes	class	NNS	I-NP	O
is	be	VBZ	B-VP	O

maximum.	maximum.	LS	B-LST	O
The	The	DT	B-NP	O
learning	learning	NN	I-NP	O
problem	problem	NN	I-NP	O
then	then	RB	B-ADVP	O
leads	lead	VBZ	B-VP	O
to	to	TO	I-VP	O
solve	solve	VB	I-VP	O
the	the	DT	B-NP	O
optimization	optimization	NN	I-NP	O
problem	problem	NN	I-NP	O

with	with	IN	B-PP	O
the	the	DT	B-NP	O
following	follow	VBG	I-NP	O
objective	objective	JJ	I-NP	O
function	function	NN	I-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


M	M	NN	B-NP	O
in	in	IN	B-PP	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein


2	2	LS	B-LST	O


d	d	NN	B-NP	O


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O
+	+	.	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


2	2	LS	B-LST	O


d	d	NN	B-NP	O


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O

p∈N1	p∈N1	NN	B-NP	O

q∈N2	q∈N2	NN	B-NP	O


2	2	LS	B-LST	O


d	d	NN	B-NP	O


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O


(	(	(	O	O
5	5	CD	B-NP	O
)	)	)	O	O


Subject	Subject	NN	B-NP	O
to	to	TO	B-PP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


4	4	LS	B-LST	O

i=1	i=1	NN	B-NP	O
T	T	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
gr	gr	NN	B-NP	B-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O
<	<	SYM	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O
i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
4	4	CD	B-NP	O
;	;	:	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
n	n	NN	B-NP	O
;	;	:	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
..	..	NN	B-NP	O
,	,	,	O	O
R	R	NN	B-NP	O
.	.	.	O	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
≥	≥	NN	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O

In	In	IN	B-PP	O
the	the	DT	B-NP	O
objective	objective	JJ	I-NP	O
function	function	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
two	two	CD	I-NP	O
first	first	JJ	I-NP	O
components	component	NNS	I-NP	O
are	be	VBP	B-VP	O
the	the	DT	B-NP	O
sum	sum	NN	I-NP	O
of	of	IN	B-PP	O
similarity	similarity	NN	B-NP	O

of	of	IN	B-PP	O
sequence	sequence	NN	B-NP	O
pairs	pair	NNS	I-NP	O
belonging	belong	VBG	B-VP	O
to	to	TO	B-PP	O
the	the	DT	B-NP	O
same	same	JJ	I-NP	O
class	class	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
last	last	JJ	I-NP	O
one	one	NN	I-NP	O
is	be	VBZ	B-VP	O
similarity	similarity	NN	B-NP	O
of	of	IN	B-PP	O

sequence	sequence	NN	B-NP	O
pairs	pair	NNS	I-NP	O
belonging	belong	VBG	B-VP	O
to	to	TO	B-PP	O
two	two	CD	B-NP	O
different	different	JJ	I-NP	O
classes	class	NNS	I-NP	O
;	;	:	O	O
d	d	NN	B-NP	O
(	(	(	O	O
x	x	NN	B-NP	O
,	,	,	I-NP	O
y	y	NN	I-NP	O
)	)	)	O	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
similarity	similarity	NN	I-NP	O
measure	measure	NN	I-NP	O

between	between	IN	B-PP	O
x	x	NN	B-NP	O
and	and	CC	I-NP	O
y	y	NN	I-NP	O
(	(	(	O	O
in	in	IN	B-PP	O
this	this	DT	B-NP	O
work	work	NN	I-NP	O
we	we	PRP	B-NP	O
use	use	VBP	B-VP	O
Euclidean	Euclidean	JJ	B-NP	O
distance	distance	NN	I-NP	O
and	and	CC	I-NP	O
L2	L2	NN	I-NP	B-DNA
norm	norm	NN	I-NP	O
)	)	)	O	O
;	;	:	O	O
N1	N1	NN	B-NP	B-protein

and	and	CC	O	O
N2	N2	NN	B-NP	B-protein
are	be	VBP	B-VP	O
the	the	DT	B-NP	O
two	two	CD	I-NP	O
index	index	NN	I-NP	O
sets	set	NNS	I-NP	O
of	of	IN	B-PP	O
high	high	JJ	B-NP	O
and	and	CC	I-NP	O
low	low	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
siRNAs	siRNA	NNS	I-NP	O
,	,	,	O	O
respectively	respectively	RB	B-ADVP	O
.	.	.	O	O

Constraints	Constraint	NNS	B-NP	O
gi	gi	NNS	I-NP	O
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O
can	can	MD	B-VP	O
also	also	RB	I-VP	O
help	help	VB	I-VP	O
to	to	TO	I-VP	O
avoid	avoid	VB	I-VP	O
the	the	DT	B-NP	O
trivial	trivial	JJ	I-NP	O
solution	solution	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
objective	objective	NN	I-NP	O

function	function	NN	B-NP	O
.	.	.	O	O


This	This	DT	B-NP	O
optimization	optimization	NN	I-NP	O
problem	problem	NN	I-NP	O
is	be	VBZ	B-VP	O
solved	solve	VBN	I-VP	O
by	by	IN	B-PP	O
the	the	DT	B-NP	O
following	follow	VBG	I-NP	O
Lagrangian	Lagrangian	JJ	I-NP	O
form	form	NN	I-NP	O


E	E	NN	B-NP	O
=	=	SYM	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein


2	2	LS	B-LST	O


d	d	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


p∈N1	p∈N1	NN	B-NP	O

q∈N1	q∈N1	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


=	=	SYM	O	O


=	=	SYM	O	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


j=1	j=1	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein

n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


+	+	SYM	O	O


λj	λj	LS	B-LST	O


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O
+	+	.	O	O


2	2	LS	B-LST	O


d	d	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O

p∈N1	p∈N1	NN	B-NP	O

q∈N2	q∈N2	NN	B-NP	O

2	2	CD	B-NP	O
−	−	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O

p∈N1	p∈N1	NN	B-NP	O

q∈N2	q∈N2	NN	B-NP	O

(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
xp	xp	VBP	B-VP	O
−	−	NN	B-NP	B-protein
xq	xq	NN	I-NP	I-protein
(	(	(	O	I-protein
cid	cid	NN	B-NP	I-protein
:	:	:	O	I-protein
3	3	CD	B-NP	I-protein
)	)	)	O	I-protein
2	2	CD	B-NP	I-protein


p∈N2	p∈N2	NN	B-NP	O

q∈N2	q∈N2	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
xp	xp	VBP	B-VP	O
−	−	NN	B-NP	B-protein
xq	xq	NN	I-NP	I-protein
(	(	(	O	I-protein
cid	cid	NN	B-NP	I-protein
:	:	:	O	I-protein
3	3	CD	B-NP	I-protein
)	)	)	O	I-protein
2	2	CD	B-NP	I-protein


2	2	CD	B-NP	O
+	+	SYM	B-NP	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


(	(	(	O	O
xp	xp	NN	B-NP	O
,	,	,	O	O
xq	xq	NN	B-NP	O
)	)	)	O	O
+	+	.	O	O


λj	λj	LS	B-LST	O


2	2	LS	B-LST	O


d	d	NN	B-NP	O


4	4	CD	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
−	−	NN	B-NP	O
1	1	CD	I-NP	O


R	R	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


+	+	SYM	O	O


μrgr	μrgr	NN	B-NP	B-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


j=1	j=1	NN	B-NP	O


i=1	i=1	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
xp	xp	VBP	B-VP	O
−	−	NN	B-NP	B-protein
xq	xq	NN	I-NP	I-protein
(	(	(	O	I-protein
cid	cid	NN	B-NP	I-protein
:	:	:	O	I-protein
3	3	CD	B-NP	I-protein
)	)	)	O	I-protein
2	2	CD	B-NP	I-protein


2	2	CD	B-NP	O
+	+	SYM	B-NP	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


λj	λj	LS	B-LST	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


4	4	CD	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


r=1	r=1	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


R	R	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
−	−	NN	B-NP	O
1	1	CD	I-NP	O


+	+	SYM	O	O


μrgr	μrgr	NN	B-NP	B-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O


j=1	j=1	NN	B-NP	O


i=1	i=1	NN	B-NP	O


r=1	r=1	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


+	+	SYM	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


j=1	j=1	NN	B-NP	O


(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O


2	2	LS	B-LST	O


2	2	LS	B-LST	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


4	4	CD	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
−	−	NN	B-NP	O
1	1	CD	I-NP	O


j=1	j=1	NN	B-NP	O


i=1	i=1	NN	B-NP	O


+	+	SYM	O	O


R	R	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


r=1	r=1	NN	B-NP	O


μrgr	μrgr	NN	B-NP	B-DNA
(	(	(	O	I-DNA
T	T	NN	B-NP	I-DNA
)	)	)	O	I-DNA
−	−	NN	B-NP	I-DNA
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O

p∈N1	p∈N1	NN	B-NP	O

q∈N2	q∈N2	NN	B-NP	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


j=1	j=1	NN	B-NP	O


(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O


2	2	LS	B-LST	O


where	where	WRB	B-ADVP	O
μr	μr	NN	B-NP	O
,	,	,	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
R	R	NN	B-NP	B-protein
and	and	CC	I-NP	O
λj	λj	NN	I-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
n	n	NN	B-NP	O
are	be	VBP	B-VP	O
Lagrangian	Lagrangian	JJ	B-NP	O
multipliers.	multipliers.	NNS	I-NP	O
To	To	TO	B-VP	O
solve	solve	VB	I-VP	O

the	the	DT	B-NP	O
minimization	minimization	NN	I-NP	O
problem	problem	NN	I-NP	O
,	,	,	O	O
an	an	DT	B-NP	O
iterative	iterative	JJ	I-NP	O
method	method	NN	I-NP	O
is	be	VBZ	B-VP	O
applied.	applied.	VBN	I-VP	O
For	For	IN	B-PP	O
each	each	DT	B-NP	O
pair	pair	NN	I-NP	O
of	of	IN	B-PP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O
,	,	,	O	O

T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
is	be	VBZ	B-VP	O
solved	solve	VBN	I-VP	O
while	while	IN	B-SBAR	O
keeping	keep	VBG	B-VP	O
the	the	DT	B-NP	O
other	other	JJ	I-NP	O
elements	element	NNS	I-NP	O
of	of	IN	B-PP	O
T	T	NN	B-NP	O
.	.	.	O	O
The	The	DT	B-NP	O
Karush-Kuhn-Tucker	Karush-Kuhn-Tucker	NNP	I-NP	O

conditions	condition	NNS	B-NP	O
are	be	VBP	B-VP	O


–	–	IN	B-PP	O
Stationarity	Stationarity	NNP	B-NP	O
:	:	:	O	O

–	–	LS	B-NP	O
Primal	Primal	JJ	I-NP	O
feasibility	feasibility	NN	I-NP	O
:	:	:	O	O
T	T	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
≥	≥	NN	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O


∂E	∂E	NN	B-NP	O


∂T	∂T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O
,	,	,	O	O
i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
4	4	CD	B-NP	O
and	and	CC	O	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
n	n	NN	B-NP	O
.	.	.	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O

4	4	LS	B-LST	O

i=1	i=1	NN	B-NP	O
T	T	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
gr	gr	NN	B-NP	B-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O
<	<	SYM	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O
i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
4	4	CD	B-NP	O
;	;	:	O	O


j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
n	n	NN	B-NP	O
;	;	:	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
...	...	:	O	O
,	,	,	O	O
R	R	NN	B-NP	O
.	.	.	O	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


405	405	CD	B-NP	O


–	–	LS	B-NP	O
Dual	Dual	JJ	I-NP	O
feasibility	feasibility	NN	I-NP	O
:	:	:	O	O
μr	μr	VB	B-VP	O
≥	≥	NN	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
R	R	NN	B-NP	O
.	.	.	O	O

–	–	LS	B-NP	O
Complementary	Complementary	JJ	I-NP	O
slackness	slackness	NN	I-NP	O
:	:	:	O	O
μrgr	μrgr	NN	B-NP	B-protein
(	(	(	O	O
T	T	NN	B-NP	O
)	)	)	O	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O
,	,	,	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
R	R	NN	B-NP	O
.	.	.	O	O


From	From	IN	B-PP	O
the	the	DT	B-NP	O
last	last	JJ	I-NP	O
three	three	CD	I-NP	O
conditions	condition	NNS	I-NP	O
,	,	,	O	O
we	we	PRP	B-NP	O
have	have	VBP	B-VP	O
μr	μr	NN	B-NP	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O
,	,	,	O	O
r	r	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
R.	R.	NNP	B-NP	O
Therefore	Therefore	NNP	I-NP	O
,	,	,	O	O
the	the	DT	O	O


stationarity	stationarity	NN	B-NP	O
condition	condition	NN	I-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
derived	derive	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O


∂E	∂E	NN	B-NP	O


∂T	∂T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O


=	=	SYM	O	O
2	2	CD	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


+2	+2	LS	B-LST	O

−2	−2	LS	B-LST	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O
(	(	(	O	O
Xp	Xp	NN	B-NP	B-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
−	−	NN	B-NP	B-protein
Xq	Xq	NN	I-NP	I-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
)	)	)	O	O

(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O
(	(	(	O	O
Xp	Xp	NN	B-NP	B-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
−	−	NN	B-NP	B-protein
Xq	Xq	NN	I-NP	I-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
)	)	)	O	O


(	(	(	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xp	Xp	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
4	4	LS	B-LST	O
)	)	)	O	O
Xq	Xq	NN	B-NP	O
[	[	(	O	O
j	j	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
]	]	)	O	O
,	,	,	O	O
T	T	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
)	)	)	O	O
(	(	(	O	O
Xp	Xp	NN	B-NP	B-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
−	−	NN	B-NP	B-protein
Xq	Xq	NN	I-NP	I-protein
[	[	(	O	I-protein
j	j	NN	B-NP	I-protein
,	,	,	O	O
i	i	NN	B-NP	O
]	]	)	O	O
)	)	)	O	O
+	+	SYM	B-NP	O
λj	λj	NN	I-NP	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O

Set	Set	NN	B-NP	O
Zp	Zp	NN	I-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
=	=	SYM	B-VP	O
(	(	(	O	O
Xp	Xp	NN	B-NP	O
−	−	NN	I-NP	O
Xq	Xq	NN	I-NP	O
)	)	)	O	O
T	T	NN	B-NP	O
and	and	CC	I-NP	O
Aij	Aij	NN	I-NP	B-protein
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
vector	vector	NN	I-NP	O
resulting	result	VBG	B-VP	O
from	from	IN	B-PP	O
the	the	DT	B-NP	O
column	column	NN	I-NP	O
j	j	NN	I-NP	O
of	of	IN	B-PP	O


p∈N1	p∈N1	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	O


matrix	matrix	NN	B-NP	O
A	A	NN	I-NP	O
by	by	IN	B-PP	O
removing	remove	VBG	B-VP	O
the	the	DT	B-NP	O
element	element	NN	I-NP	O
A	A	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
.	.	.	O	O
Therefore	Therefore	RB	B-ADVP	O
,	,	,	O	O
the	the	DT	B-NP	O
above	above	JJ	I-NP	O
formulation	formulation	NN	I-NP	O
is	be	VBZ	B-VP	O

derived	derive	VBN	B-VP	O
as	as	IN	B-PP	O
follows	follow	VBZ	B-VP	O


∂E	∂E	NN	B-NP	O


∂T	∂T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O


=	=	SYM	O	O
2	2	CD	B-NP	O
(	(	(	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
+	+	SYM	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein

−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


p∈N1	p∈N1	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


+2T	+2T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein


2	2	LS	B-LST	O

p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
+	+	SYM	B-NP	O


Z	Z	NN	B-NP	B-protein


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


2	2	LS	B-LST	O


Z	Z	NN	B-NP	B-protein


p∈N1	p∈N1	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O

p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O


Z	Z	NN	B-NP	B-protein


+	+	SYM	O	O
λj	λj	NN	B-NP	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O


We	We	PRP	B-NP	O
define	define	VBP	B-VP	O
the	the	DT	B-NP	O
following	follow	VBG	I-NP	O
equations	equation	NNS	I-NP	O


S	S	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O


2	2	LS	B-LST	O

p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
+	+	SYM	B-NP	O


Z	Z	NN	B-NP	B-protein


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


B	B	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
+	+	SYM	B-NP	O


p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


2	2	LS	B-LST	O


Z	Z	NN	B-NP	B-protein


2	2	LS	B-LST	O

p	p	NN	B-NP	O
,	,	,	I-NP	O
q	q	NN	I-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O


Z	Z	NN	B-NP	B-protein


p∈N1	p∈N1	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


p	p	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	B-protein


p	p	NN	B-NP	O
,	,	,	O	O
q∈N1	q∈N1	NN	B-NP	B-protein

−	−	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O

p∈N1	p∈N1	NN	B-NP	O
,	,	,	O	O
q∈N2	q∈N2	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O
(	(	(	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
)	)	)	O	O
ij	ij	NN	B-NP	O
,	,	,	O	O
Tij	Tij	NN	B-NP	B-protein
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
5	5	LS	B-LST	O
)	)	)	O	O
Zp	Zp	NN	B-NP	B-DNA
,	,	,	O	O
q	q	NN	B-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
]	]	)	O	O
.	.	.	O	O


Substitute	Substitute	NN	B-NP	O
(	(	(	O	O
6	6	CD	B-NP	O
)	)	)	O	O
and	and	CC	O	O
(	(	(	O	O
7	7	CD	B-NP	O
)	)	)	O	O
to	to	TO	O	O


∂E	∂E	NN	B-NP	O


∂T	∂T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
,	,	,	O	O
we	we	PRP	B-NP	O
have	have	VBP	B-VP	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
=	=	SYM	B-NP	O


−λj	−λj	LS	B-LST	O

2	2	LS	B-LST	O


−	−	NN	B-NP	O
B	B	NN	I-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


(	(	(	O	O
6	6	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
7	7	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O


At	At	IN	B-PP	O
a	a	DT	B-NP	O
column	column	NN	I-NP	O
j	j	NN	I-NP	O
,	,	,	O	O
T	T	NN	B-NP	O
has	have	VBZ	B-VP	O
to	to	TO	I-VP	O
satisfy	satisfy	VB	I-VP	O

−λj	−λj	LS	B-LST	O

2	2	LS	B-LST	O


i1=1	i1=1	NN	B-NP	O
T	T	NN	I-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
⇔	⇔	NN	I-NP	O
4	4	CD	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O


i1=1	i1=1	NN	B-NP	O


−	−	NN	B-NP	O
B	B	NN	I-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


=	=	SYM	O	O
1	1	CD	B-NP	O
⇒	⇒	NN	I-NP	O
−λj	−λj	.	O	O


2	2	LS	B-LST	O


=	=	SYM	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O


1	1	CD	B-NP	O
+	+	SYM	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
2	2	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O


i1=1	i1=1	NN	B-NP	O


B	B	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

1	1	CD	B-NP	O


(	(	(	O	O
9	9	CD	B-NP	O
)	)	)	O	O


i1=1	i1=1	NN	B-NP	O


S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


Substitute	Substitute	NN	B-NP	O
(	(	(	O	O
9	9	CD	B-NP	O
)	)	)	O	O
to	to	TO	B-PP	O
(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
equation	equation	NN	B-NP	O
(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
derived	derive	VBN	I-VP	O
as	as	IN	B-SBAR	O


1+	1+	LS	B-LST	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
4	4	CD	B-NP	O


i1=1	i1=1	NN	B-NP	O


i1=1	i1=1	NN	B-NP	O


B	B	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


1	1	CD	B-NP	O


S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


T	T	NN	B-NP	O
[	[	(	I-NP	O
i	i	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
=	=	SYM	B-NP	O


−B	−B	NN	B-NP	B-protein
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


1+	1+	LS	B-LST	O


=	=	SYM	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O


B	B	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O
−B	−B	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


i1	i1	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
=i	=i	.	O	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O


4	4	LS	B-LST	O

i1	i1	NN	B-NP	B-protein
=1	=1	CD	I-NP	I-protein


S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i	i	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O

S	S	NN	B-NP	O
(	(	(	O	O
i1	i1	NN	B-NP	O
,	,	,	O	O
j	j	NN	B-NP	O
)	)	)	O	O


(	(	(	O	O
10	10	CD	B-NP	O
)	)	)	O	O


406	406	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


In	In	IN	B-PP	O
this	this	DT	B-NP	O
task	task	NN	I-NP	O
,	,	,	O	O
K	K	NN	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
are	be	VBP	B-VP	O
used	use	VBN	I-VP	O
to	to	TO	B-VP	O
learn	learn	VB	I-VP	O
K	K	NN	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
.	.	.	O	O

The	The	DT	B-NP	O
main	main	JJ	I-NP	O
steps	step	NNS	I-NP	O
are	be	VBP	B-VP	O
summarized	summarize	VBN	I-VP	O
in	in	IN	B-PP	O
Algorithm	Algorithm	NNP	B-NP	O
1.	1.	CD	I-NP	O
For	For	IN	B-PP	O
each	each	DT	B-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O

the	the	DT	B-NP	O
algorithm	algorithm	NN	I-NP	O
will	will	MD	B-VP	O
update	update	VB	I-VP	O
each	each	DT	B-NP	O
element	element	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
according	accord	VBG	B-VP	O

to	to	TO	B-PP	O
equation	equation	NN	B-NP	O
(	(	(	O	O
10	10	CD	B-NP	O
)	)	)	O	O
.	.	.	O	O
In	In	IN	B-PP	O
each	each	DT	B-NP	O
iterative	iterative	JJ	I-NP	O
step	step	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
without	without	IN	B-PP	O
trick	trick	NN	B-NP	O

inequality	inequality	NN	B-NP	O
constraints	constraint	NNS	I-NP	O
is	be	VBZ	B-VP	O
updated	update	VBN	I-VP	O
to	to	TO	I-VP	O
reach	reach	VB	I-VP	O
the	the	DT	B-NP	O
global	global	JJ	I-NP	O
optimal	optimal	JJ	I-NP	O
solution.	solution.	NN	I-NP	O
If	If	IN	B-SBAR	O
updated	update	VBN	O	O

elements	element	NNS	B-NP	O
in	in	IN	B-PP	O
a	a	DT	B-NP	O
column	column	NN	I-NP	O
satisfy	satisfy	VBP	B-VP	O
the	the	DT	B-NP	O
trick	trick	NN	I-NP	O
inequality	inequality	NN	I-NP	O
constraints	constraint	NNS	I-NP	O
characterizing	characterize	VBG	B-VP	O
the	the	DT	B-NP	O

condition	condition	NN	B-NP	O
at	at	IN	B-PP	O
the	the	DT	B-NP	O
corresponding	corresponding	JJ	I-NP	O
position	position	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
that	that	DT	B-NP	O
column	column	NN	I-NP	O
will	will	MD	B-VP	O
be	be	VB	I-VP	O
updated	update	VBN	I-VP	O

to	to	TO	B-PP	O
the	the	DT	B-NP	O
target	target	NN	I-NP	O
solution.	solution.	RB	B-ADVP	O
The	The	DT	B-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
is	be	VBZ	B-VP	O
updated	update	VBN	I-VP	O
until	until	IN	B-PP	O
meeting	meet	VBG	B-VP	O
the	the	DT	B-NP	O


convergence	convergence	NN	B-NP	O
criteria.	criteria.	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O
.	.	.	O	O
(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O
F	F	NN	B-NP	B-protein
ro	ro	NN	I-NP	I-protein
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
Frobenious	Frobenious	JJ	I-NP	O
norm	norm	NN	I-NP	O
of	of	IN	B-PP	O
a	a	DT	B-NP	O
matrix	matrix	NN	I-NP	O
.	.	.	O	O


Algorithm	Algorithm	NNP	B-NP	O
1.	1.	CD	I-NP	O
Transformation	Transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
learning	learn	VBG	B-VP	O


Input	Input	NN	B-NP	O
:	:	:	O	O
A	A	DT	B-NP	O
data	datum	NNS	I-NP	O
set	set	VBP	B-VP	O
S	S	NN	B-NP	O
=	=	SYM	B-VP	O
{	{	NN	B-NP	O
(	(	(	O	O
sl	sl	NN	B-NP	O
,	,	,	O	O
yl	yl	NN	B-NP	O
)	)	)	O	O
}	}	NN	B-NP	O
N	N	NN	I-NP	O

labels	label	NNS	B-NP	O
,	,	,	O	O
a	a	DT	B-NP	O
set	set	VBN	I-NP	O
DR	DR	NN	I-NP	O
of	of	IN	B-PP	O
K	K	NN	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
length	length	NN	I-NP	O
n	n	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
.	.	.	O	O

Output	Output	NN	B-NP	O
:	:	:	O	O
K	K	NN	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
T1	T1	NN	I-NP	O
,	,	,	O	O
T2	T2	NN	B-NP	B-DNA
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
TK	TK	NN	B-NP	O
.	.	.	O	O

Encoding	Encode	VBG	B-VP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
in	in	IN	B-PP	O
S	S	NN	B-NP	O
.	.	.	O	O

for	for	IN	B-PP	O
rulek	rulek	NN	B-NP	O
in	in	IN	B-PP	O
DR	DR	NN	B-NP	O
do	do	VBP	B-VP	O


1	1	CD	B-NP	O
where	where	WRB	B-ADVP	O
sl	sl	NN	B-NP	O
are	be	VBP	B-VP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
and	and	CC	O	O
yl	yl	NN	B-NP	O
are	be	VBP	B-VP	O
their	their	PRP$	B-NP	O


Form	Form	VB	B-VP	O
the	the	DT	B-NP	O
set	set	NN	I-NP	O
of	of	IN	B-PP	O
constraints	constraint	NNS	B-NP	O
Ck	Ck	NN	I-NP	B-protein
based	base	VBN	B-VP	O
on	on	IN	B-PP	O
rulek	rulek	NN	B-NP	O

Initialize	Initialize	VB	B-VP	O
the	the	DT	B-NP	O
transformation	transformation	NN	I-NP	B-protein
matrix	matrix	NN	I-NP	I-protein
Tk	Tk	NN	I-NP	I-protein
satisfying	satisfy	VBG	B-VP	O
Ck	Ck	NNP	B-NP	O
.	.	.	O	O

t	t	NN	B-NP	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O
{	{	NN	I-NP	O
Iterative	Iterative	JJ	I-NP	O
step	step	NN	I-NP	O
}	}	.	O	O

repeat	repeat	NN	B-NP	O


t	t	NN	B-NP	O
←	←	NN	I-NP	O
t	t	NN	I-NP	O
+	+	SYM	B-NP	O
1	1	CD	I-NP	O

for	for	IN	B-PP	O
j	j	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
to	to	TO	B-PP	O
n	n	NN	B-NP	O
do	do	VB	B-VP	O


(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O

k	k	NN	B-NP	O


[	[	(	O	O
.	.	NN	B-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
{	{	VBP	B-VP	O
A	A	DT	B-NP	O
temporary	temporary	JJ	I-NP	O
vector	vector	NN	I-NP	O
}	}	.	O	O


v	v	NN	B-NP	O
=	=	SYM	B-VP	O
T	T	NN	B-NP	O

for	for	IN	B-PP	O
i	i	NN	B-NP	O
=	=	SYM	B-VP	O
1	1	CD	B-NP	O
to	to	TO	B-PP	O
4	4	CD	B-NP	O
do	do	NN	I-NP	O


Compute	Compute	NN	B-NP	O
v	v	NN	I-NP	O
[	[	(	O	O
i	i	NN	B-NP	O
]	]	)	O	O
using	use	VBG	B-VP	O
equation	equation	NN	B-NP	O
(	(	(	O	O
10	10	CD	B-NP	O
)	)	)	O	O


end	end	VB	B-VP	O
for	for	IN	B-PP	O

if	if	IN	B-SBAR	O
(	(	(	O	O
v	v	NN	B-NP	O
satisfies	satisfie	VBZ	B-VP	O
the	the	DT	B-NP	O
constraints	constraint	NNS	I-NP	O
at	at	IN	B-PP	O
the	the	DT	B-NP	O
position	position	NN	I-NP	O
j	j	NN	I-NP	O
in	in	IN	B-PP	O
Ck	Ck	NN	B-NP	O
)	)	)	O	O
then	then	RB	O	O


(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O


k	k	NN	B-NP	O
[	[	(	I-NP	O
.	.	NN	I-NP	O
,	,	,	I-NP	O
j	j	NN	I-NP	O
]	]	)	O	O
←	←	NN	B-NP	O
v	v	NN	I-NP	O


T	T	NN	B-NP	O


end	end	VB	B-VP	O
if	if	IN	B-SBAR	O

end	end	VB	B-VP	O
for	for	IN	B-PP	O

(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
T	T	NN	B-NP	O
(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O

k	k	NN	B-NP	O


until	until	IN	B-SBAR	O
(	(	(	O	O


−T	−T	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
F	F	NN	B-NP	O
ro	ro	NN	I-NP	O


k	k	NN	B-NP	O


(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
T	T	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O


k	k	NN	B-NP	O


(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O
F	F	NN	B-NP	O
ro	ro	NN	I-NP	O


≤	≤	LS	B-LST	O

	
	:	O	O
)	)	)	O	O
or	or	CC	O	O
(	(	(	O	O
t	t	NN	B-NP	O
>	>	SYM	O	O
tM	tM	NN	B-NP	O
ax	ax	NN	I-NP	O
)	)	)	O	O


end	end	VB	B-VP	O
for	for	IN	B-PP	O


2.2	2.2	CD	B-NP	O
Tensor	Tensor	NNP	I-NP	O
Regression	Regression	NNP	I-NP	O
Model	Model	NNP	I-NP	O
Learning	Learning	NNP	I-NP	O


Given	Give	VBN	B-VP	O
a	a	DT	B-NP	O
siRNA	siRNA	NN	I-NP	O
data	datum	NNS	I-NP	O
set	set	VBP	B-VP	O
D	D	NN	B-NP	O
=	=	SYM	B-VP	O
{	{	NN	B-NP	O
(	(	(	O	O
sl	sl	NN	B-NP	O
,	,	,	O	O
yl	yl	NN	B-NP	O
)	)	)	O	O
}	}	NN	B-NP	O
N	N	NN	I-NP	O

1	1	CD	B-NP	O
where	where	WRB	B-ADVP	O
sl	sl	NN	B-NP	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
lth	lth	NN	I-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequence	sequence	NN	I-NP	I-DNA

of	of	IN	B-PP	O
size	size	NN	B-NP	O
n	n	NN	I-NP	O
and	and	CC	O	O
yl	yl	NN	B-NP	B-protein
∈	∈	NN	I-NP	I-protein
R	R	NN	I-NP	I-protein
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
score	score	NN	I-NP	O
of	of	IN	B-PP	O
sl.	sl.	NN	B-NP	O
Let	Let	NNP	I-NP	O
Xl	Xl	NNP	I-NP	O
denotes	denote	VBZ	B-VP	O
the	the	DT	O	O

encoding	encode	VBG	B-VP	O
matrix	matrix	NN	B-NP	O
of	of	IN	B-PP	O
sl.	sl.	NN	B-NP	O
Each	Each	DT	B-NP	O
encoding	encode	VBG	I-NP	O
matrix	matrix	NN	I-NP	B-protein
X	X	NN	I-NP	I-protein
is	be	VBZ	B-VP	O
transformed	transform	VBN	I-VP	O
to	to	TO	B-PP	O
K	K	NN	B-NP	O
represen-	represen-	NN	I-NP	O

tations	tation	NNS	B-NP	O
by	by	IN	B-PP	O
K	K	NN	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
,	,	,	O	O
(	(	(	O	O
T1	T1	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
T2	T2	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
TK	TK	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
)	)	)	O	O
.	.	.	O	O
R	R	NN	B-NP	B-protein
(	(	(	O	O
X	X	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O

(	(	(	O	O
T1	T1	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
T2	T2	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
TK	TK	NN	B-NP	O
◦	◦	NN	I-NP	O
X	X	NN	I-NP	O
)	)	)	O	O
T	T	NN	B-NP	O
denotes	denote	VBZ	B-VP	O
the	the	DT	B-NP	O
second	second	JJ	I-NP	O
order	order	NN	I-NP	O
tensor	tensor	NN	I-NP	O
of	of	IN	B-PP	O
size	size	NN	B-NP	O
K	K	NN	I-NP	O
×	×	NN	I-NP	O
n	n	NN	I-NP	O
.	.	.	O	O


The	The	DT	B-NP	O
regression	regression	NN	I-NP	O
model	model	NN	I-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
defined	define	VBN	I-VP	O
as	as	IN	B-SBAR	O
the	the	DT	B-NP	O
following	follow	VBG	I-NP	O
bilinear	bilinear	JJ	I-NP	O
form	form	NN	I-NP	O


f	f	NN	B-NP	O
(	(	(	I-NP	O
x	x	NN	I-NP	O
)	)	)	I-NP	O
=	=	JJ	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
X	X	NN	B-NP	O
)	)	)	O	O
β	β	.	O	O


(	(	(	O	O
11	11	CD	B-NP	O
)	)	)	O	O


where	where	WRB	B-ADVP	O
α	α	NN	B-NP	O
=	=	SYM	B-NP	O
(	(	(	O	O
α1	α1	NN	B-NP	O
,	,	,	O	O
α2	α2	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
αK	αK	NN	B-NP	B-protein
)	)	)	O	O
is	be	VBZ	B-VP	O
a	a	DT	B-NP	O
weight	weight	NN	I-NP	O
vector	vector	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
K	K	NN	I-NP	O
representations	representation	NNS	I-NP	O
of	of	IN	B-PP	O
X	X	NN	B-NP	B-DNA
and	and	CC	O	O

β	β	NN	B-NP	O
=	=	SYM	B-NP	O
(	(	(	O	O
β1	β1	NN	B-NP	O
,	,	,	O	O
β2	β2	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
βn	βn	LS	B-LST	O
)	)	)	O	O
T	T	NN	B-NP	O
is	be	VBZ	B-VP	O
a	a	DT	B-NP	O
parameter	parameter	NN	I-NP	O
vector	vector	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
model	model	NN	I-NP	O
,	,	,	O	O
and	and	CC	O	O
αR	αR	NN	B-NP	B-protein
(	(	(	O	I-protein
X	X	NN	B-NP	I-protein
)	)	)	O	I-protein
component	component	NN	B-NP	I-protein

is	be	VBZ	O	O
the	the	DT	B-NP	O
linear	linear	JJ	I-NP	O
combination	combination	NN	I-NP	O
of	of	IN	B-PP	O
representations	representation	NNS	B-NP	O
T1	T1	NN	I-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
T2	T2	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
TK	TK	NNP	B-NP	O
◦	◦	NNP	I-NP	O
X.	X.	NNP	I-NP	O
It	It	PRP	B-NP	O
also	also	RB	B-ADVP	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


407	407	CD	B-NP	O


shows	show	VBZ	B-VP	O
the	the	DT	B-NP	O
relationship	relationship	NN	I-NP	B-DNA
among	among	IN	B-PP	I-DNA
elements	element	NNS	B-NP	I-DNA
on	on	IN	B-PP	O
each	each	DT	B-NP	O
column	column	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
second	second	JJ	I-NP	O
order	order	NN	I-NP	O
tensor	tensor	NN	I-NP	O


or	or	CC	O	O
each	each	DT	B-NP	O
dimension	dimension	NN	I-NP	O
of	of	IN	B-PP	O
Tk	Tk	NN	B-NP	B-protein
◦	◦	NN	I-NP	I-protein
X	X	NN	I-NP	I-protein
,	,	,	O	O
k	k	NN	B-NP	O
=	=	SYM	B-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
2	2	CD	I-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
Equation	Equation	NNP	I-NP	O
(	(	(	O	O
11	11	CD	B-NP	O
)	)	)	O	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
derived	derive	VBN	I-VP	O
as	as	IN	B-SBAR	O


follows	follow	VBZ	B-VP	O


f	f	NN	B-NP	O
(	(	(	I-NP	O
X	X	NN	I-NP	O
)	)	)	I-NP	O
=	=	JJ	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
X	X	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
=	=	SYM	B-VP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O


β	β	LS	B-LST	O
⊗	⊗	:	O	O
α	α	NN	B-NP	O


T	T	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADVP	O
:	:	:	O	O
7	7	CD	O	O
)	)	)	O	O
T	T	NN	B-NP	O


vec	vec	NN	B-NP	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
X	X	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O
=	=	.	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADVP	O
:	:	:	O	O
7	7	CD	B-NP	O
)	)	)	O	O

T	T	NN	B-NP	O
⊗	⊗	VBD	B-VP	O
α	α	.	O	O


β	β	LS	B-LST	O


vec	vec	NN	B-NP	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
X	X	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
12	12	CD	B-NP	O
)	)	)	O	O


where	where	WRB	B-ADVP	O
A	A	DT	B-NP	O
⊗	⊗	NN	I-NP	B-DNA
B	B	NN	I-NP	I-DNA
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
Kronecker	Kronecker	NNP	I-NP	O
product	product	NN	I-NP	O
of	of	IN	B-PP	O
two	two	CD	B-NP	O
matrices	matrix	NNS	I-NP	O
A	A	NN	I-NP	O
and	and	CC	I-NP	O
B	B	NN	I-NP	O
,	,	,	O	O
and	and	CC	O	O
vec	vec	NN	B-NP	O
(	(	(	O	O
A	A	NN	B-NP	O
)	)	)	O	O
is	be	VBZ	B-VP	O

the	the	DT	B-NP	O
vectorization	vectorization	NN	I-NP	O
of	of	IN	B-PP	O
matrix	matrix	NN	B-NP	O
A.	A.	NN	I-NP	O
The	The	DT	B-NP	O
weight	weight	NN	I-NP	O
vector	vector	NN	I-NP	O
α	α	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
parameter	parameter	NN	I-NP	O
vector	vector	NN	I-NP	O
β	β	.	O	O

are	be	VBP	B-VP	O
learned	learn	VBN	I-VP	O
by	by	IN	B-PP	O
minimizing	minimize	VBG	B-VP	O
the	the	DT	B-NP	O
following	follow	VBG	I-NP	O
regularized	regularize	VBN	I-NP	O
risk	risk	NN	I-NP	O
function	function	NN	I-NP	O


L	L	NN	B-NP	O
(	(	(	O	O
α	α	NN	B-NP	O
,	,	,	O	O
β	β	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	.	O	O


T	T	NN	B-NP	O
⊗	⊗	NN	I-NP	O
α	α	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


F	F	NN	B-NP	B-protein
ro	ro	NN	I-NP	I-protein


(	(	(	O	O
13	13	CD	B-NP	O
)	)	)	O	O


where	where	WRB	B-ADVP	O
λ	λ	NN	B-NP	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
turning	turn	VBG	I-NP	O
parameter	parameter	NN	I-NP	O
to	to	TO	B-PP	O
tradeoff	tradeoff	NN	B-NP	O
between	between	IN	B-PP	O
bias	bias	NN	B-NP	O
and	and	CC	I-NP	O
variance	variance	NN	I-NP	O
,	,	,	O	O
and	and	CC	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O
βT	βT	NN	B-NP	O
⊗	⊗	NN	I-NP	O
α	α	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O
F	F	NN	B-NP	B-protein
ro	ro	NN	I-NP	I-protein
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
Frobenius	Frobenius	NNP	I-NP	O
norm	norm	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
first	first	JJS	I-NP	O
order	order	NN	I-NP	O
tensor	tensor	NN	I-NP	O
βT	βT	NN	I-NP	O
⊗	⊗	FW	B-NP	O
α.	α.	FW	I-NP	O
L	L	NN	I-NP	O
(	(	(	O	O
α	α	NN	B-NP	O
,	,	,	O	O
β	β	NN	B-NP	O
)	)	)	O	O


can	can	MD	B-VP	O
be	be	VB	I-VP	O
derived	derive	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O


L	L	NN	B-NP	O
(	(	(	O	O
α	α	NN	B-NP	O
,	,	,	O	O
β	β	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O


(	(	(	B-LST	O
αkβj	αkβj	LS	I-LST	O
)	)	)	O	O


=	=	SYM	O	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O

N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


=	=	SYM	O	O


l=1	l=1	NN	B-NP	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O


K	K	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


2	2	LS	B-LST	O

j	j	NN	B-NP	O


β	β	LS	B-LST	O


2	2	LS	B-LST	O

k	k	NN	B-NP	O


α	α	LS	B-LST	O


k=1	k=1	NN	B-NP	O


j=1	j=1	NN	B-NP	O


n	n	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


j=1	j=1	NN	B-NP	O


K	K	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O

K	K	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


k=1	k=1	NN	B-NP	O


α	α	LS	B-LST	O


k=1	k=1	NN	B-NP	O


2	2	LS	B-LST	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
2	2	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O


2	2	LS	B-LST	O


k	k	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O

2=	2=	LS	B-LST	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


2	2	LS	B-LST	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


2	2	CD	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


2	2	CD	B-NP	O
(	(	(	O	O
14	14	CD	B-NP	O
)	)	)	O	O


The	The	DT	B-NP	O
risk	risk	NN	I-NP	O
function	function	NN	I-NP	O
with	with	IN	B-PP	O
Frobenius	Frobenius	NN	B-NP	O
norm	norm	NN	I-NP	O
is	be	VBZ	B-VP	O
converted	convert	VBN	I-VP	O
to	to	TO	B-PP	O
equation	equation	NN	B-NP	O
(	(	(	O	O
14	14	CD	B-NP	O
)	)	)	O	O
with	with	IN	B-PP	O

L2	L2	NN	B-NP	B-DNA
norm.	norm.	RB	B-ADVP	O
In	In	IN	B-SBAR	O
order	order	NN	O	O
to	to	TO	B-VP	O
solve	solve	VB	I-VP	O
this	this	DT	B-NP	O
optimization	optimization	NN	I-NP	O
problem	problem	NN	I-NP	O
,	,	,	O	O
an	an	DT	B-NP	O
alternative	alternative	JJ	I-NP	O
iteration	iteration	NN	I-NP	O

method	method	NN	B-NP	O
is	be	VBZ	B-VP	O
used.	used.	VBN	I-VP	O
At	At	IN	B-PP	O
each	each	DT	B-NP	O
iteration	iteration	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
parameter	parameter	NN	I-NP	O
vector	vector	NN	I-NP	O
β	β	NN	I-NP	O
is	be	VBZ	B-VP	O
effectively	effectively	RB	I-VP	O
solved	solve	VBN	I-VP	O

by	by	IN	B-PP	O
keeping	keep	VBG	B-VP	O
the	the	DT	B-NP	O
weight	weight	NN	I-NP	O
vector	vector	NN	I-NP	O
α	α	NN	I-NP	O
and	and	CC	O	O
vice	vice	NN	B-NP	O
versa	versa	NN	I-NP	O
.	.	.	O	O

(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


+	+	SYM	O	O
2λα	2λα	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


∂α	∂α	NN	B-NP	O
=	=	SYM	B-VP	O
−2	−2	CD	B-NP	O


2=	2=	CD	B-NP	O
0	0	CD	I-NP	O


∂L	∂L	NN	B-NP	O
(	(	(	O	O
α	α	NN	B-NP	O
,	,	,	O	O
β	β	NN	B-NP	O
)	)	)	O	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


T	T	NN	B-NP	O


⇔	⇔	LS	B-LST	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


l=1	l=1	NN	B-NP	O


α	α	NN	B-NP	O
(	(	(	O	O
R	R	NN	B-NP	B-protein
(	(	(	O	I-protein
Xl	Xl	NN	B-NP	I-protein
)	)	)	O	I-protein
β	β	NN	B-NP	I-protein
)	)	)	O	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


⇒	⇒	SYM	O	O
α	α	NN	B-NP	O
=	=	SYM	B-VP	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


yl	yl	NN	B-NP	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


T	T	NN	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


T	T	NN	B-NP	O
−	−	NN	I-NP	O
N	N	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


yl	yl	NN	B-NP	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


T	T	NN	B-NP	O


+	+	SYM	O	O
λα	λα	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−1	−1	.	O	O


2=	2=	CD	B-NP	O
0	0	CD	I-NP	O


(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
R	R	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O


T	T	NN	B-NP	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O

2	2	CD	B-NP	O
I	I	PRP	B-NP	O


(	(	(	O	O
15	15	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
yl	yl	NN	B-NP	O
−	−	NN	I-NP	O
αR	αR	NN	I-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
β	β	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


T	T	NN	B-NP	O


+	+	SYM	O	O
2λβ	2λβ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


2=	2=	CD	B-NP	O
0	0	CD	I-NP	O


T	T	NN	B-NP	O
−	−	NN	I-NP	O
N	N	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


yl	yl	NN	B-NP	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


T	T	NN	B-NP	O


+	+	SYM	O	O
λβ	λβ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


2=	2=	CD	B-NP	O
0	0	CD	I-NP	O


∂L	∂L	NN	B-NP	O
(	(	(	O	O
α	α	NN	B-NP	O
,	,	,	O	O
β	β	NN	B-NP	O
)	)	)	O	O


∂β	∂β	NN	B-NP	O
=	=	SYM	B-VP	O
−2	−2	CD	B-NP	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


⇔	⇔	LS	B-LST	O


⇔	⇔	LS	B-LST	O


αR	αR	NN	B-NP	B-protein
(	(	(	O	I-protein
Xl	Xl	NN	B-NP	I-protein
)	)	)	O	I-protein
β	β	NN	B-NP	I-protein
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O

N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADVP	O
:	:	:	O	O
7	7	CD	B-NP	O
)	)	)	O	O


T	T	NN	B-NP	O
⊗	⊗	NN	I-NP	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


⇒	⇒	SYM	O	O
β	β	NN	B-NP	O
=	=	SYM	B-VP	O


(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


T	T	NN	B-NP	O
⊗	⊗	NN	I-NP	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADVP	O
:	:	:	O	O
7	7	CD	B-NP	O
)	)	)	O	O


+	+	SYM	O	O
λ	λ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O

2	2	CD	B-NP	O
I	I	PRP	B-NP	O


l=1	l=1	NN	B-NP	O

(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
4	4	CD	B-NP	O
)	)	)	O	O


N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
6	6	CD	B-NP	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


β	β	LS	B-LST	O
−	−	:	O	O
N	N	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


yl	yl	NN	B-NP	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


T	T	NN	B-NP	O


+	+	SYM	O	O
λβ	λβ	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


2=	2=	CD	B-NP	O
0	0	CD	I-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
5	5	CD	B-NP	O
)	)	)	O	O
−1	−1	CD	B-NP	O
N	N	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


l=1	l=1	NN	B-NP	O


T	T	NN	B-NP	O

yl	yl	NN	B-NP	O
(	(	(	O	O
αR	αR	NN	B-NP	O
(	(	(	O	O
Xl	Xl	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


(	(	(	O	O
16	16	CD	B-NP	O
)	)	)	O	O


408	408	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


Our	Our	PRP$	B-NP	O
proposed	propose	VBN	I-NP	O
tensor	tensor	NN	I-NP	O
regression	regression	NN	I-NP	O
model	model	NN	I-NP	O
learning	learning	NN	I-NP	O
is	be	VBZ	B-VP	O
summarized	summarize	VBN	I-VP	O
in	in	IN	B-PP	O
Algorithm	Algorithm	NN	B-NP	B-protein
2	2	CD	I-NP	I-protein
.	.	.	O	O

In	In	IN	B-PP	O
this	this	DT	B-NP	O
algorithm	algorithm	NN	I-NP	O
,	,	,	O	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
are	be	VBP	B-VP	O
firstly	firstly	RB	I-VP	O
represented	represent	VBN	I-VP	O
as	as	IN	B-PP	O
encoding	encode	VBG	B-VP	O
matrices	matrix	NNS	B-NP	O
.	.	.	O	O

The	The	DT	B-NP	O
encoding	encode	VBG	I-NP	O
matrices	matrix	NNS	I-NP	O
are	be	VBP	B-VP	O
then	then	RB	I-VP	O
transformed	transform	VBN	I-VP	O
to	to	TO	B-PP	O
tensors	tensor	NNS	B-NP	O
by	by	IN	B-PP	O
using	use	VBG	B-VP	O
K	K	NN	B-NP	O
transforma-	transforma-	NN	I-NP	O

tion	tion	NN	B-NP	O
matrices.	matrices.	NN	I-NP	O
After	After	IN	B-PP	O
that	that	DT	B-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
weight	weight	NN	I-NP	O
vector	vector	NN	I-NP	O
α	α	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
coefficient	coefficient	NN	I-NP	O
vector	vector	NN	I-NP	O
β	β	NN	I-NP	O
are	be	VBP	B-VP	O

updated	update	VBN	B-VP	O
until	until	IN	B-PP	O
meeting	meet	VBG	B-VP	O
the	the	DT	B-NP	O
convergence	convergence	NN	I-NP	O
criteria	criterion	NNS	I-NP	O
,	,	,	O	O
where	where	WRB	B-ADVP	O
tMax	tMax	NN	B-NP	B-protein
denotes	denote	VBZ	B-VP	O
the	the	DT	B-NP	O
maxi-	maxi-	NN	I-NP	O

mum	mum	JJ	B-NP	O
iterative	iterative	JJ	I-NP	O
step	step	NN	I-NP	O
to	to	TO	B-VP	O
update	update	VB	I-VP	O
α	α	NN	B-NP	O
and	and	CC	I-NP	O
β	β	NN	I-NP	O
,	,	,	O	O
and	and	CC	O	O

1	
1	NN	B-NP	O
and	and	CC	I-NP	O

2	
2	NN	I-NP	O
are	be	VBP	B-VP	O
thresholds	threshold	NNS	B-NP	O
for	for	IN	B-PP	O
vectors	vector	NNS	B-NP	O

α	α	NN	B-NP	O
and	and	CC	I-NP	O
β	β	NN	I-NP	O
.	.	.	O	O


Algorithm	Algorithm	NNP	B-NP	O
2.	2.	CD	I-NP	O
Tensor	Tensor	NNP	I-NP	O
Regression	Regression	NNP	I-NP	O
Model	Model	NNP	I-NP	O
Learning	Learning	NNP	I-NP	O


Input	Input	NN	B-NP	O
:	:	:	O	O
A	A	DT	B-NP	O
data	datum	NNS	I-NP	O
set	set	VBP	B-VP	O
S	S	NN	B-NP	O
=	=	SYM	B-VP	O
{	{	NN	B-NP	O
(	(	(	O	O
si	si	NN	B-NP	O
,	,	,	O	O
yi	yi	NN	B-NP	O
)	)	)	O	O
}	}	NN	B-NP	O
N	N	NN	I-NP	O

K	K	NN	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
R1	R1	NN	I-NP	B-protein
,	,	,	O	O
R2	R2	NN	B-NP	B-protein
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
Rk	Rk	NNP	B-NP	O
,	,	,	O	O
and	and	CC	O	O
the	the	DT	B-NP	O
length	length	NN	I-NP	O
n	n	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA
.	.	.	O	O

Output	Output	NN	B-NP	O
:	:	:	O	O
Weight	Weight	NN	B-NP	O
vector	vector	NN	I-NP	O
α	α	NN	I-NP	O
=	=	SYM	B-VP	O
(	(	(	O	O
α1	α1	NN	B-NP	O
,	,	,	O	O
α2	α2	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
αk	αk	VBP	B-VP	O
)	)	)	O	O
and	and	CC	O	O
parameter	parameter	NN	B-NP	O
vector	vector	NN	I-NP	O

β	β	NN	B-NP	O
=	=	SYM	B-NP	O
(	(	(	O	O
β1	β1	NN	B-NP	O
,	,	,	O	O
β2	β2	NN	B-NP	O
,	,	,	O	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
,	,	,	O	O
βn	βn	NN	B-NP	O
)	)	)	O	O
that	that	WDT	B-NP	O
minimize	minimize	VBP	B-VP	O
the	the	DT	B-NP	O
regularized	regularize	VBN	I-NP	O
risk	risk	NN	I-NP	O
function	function	NN	I-NP	O


1	1	CD	B-NP	O
where	where	WRB	B-ADVP	O
si	si	NNS	B-NP	O
are	be	VBP	B-VP	O
scoring	score	VBG	I-VP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
and	and	CC	O	O
yi	yi	NNS	B-NP	O
∈	∈	NNP	B-NP	O
R	R	NNP	I-NP	O
.	.	.	O	O


–	–	LS	B-NP	B-DNA
Represent	Represent	NNP	I-NP	I-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
in	in	IN	B-PP	O
S	S	NN	B-NP	O
as	as	IN	B-PP	O
enconding	enconde	VBG	B-VP	O
matrices	matrix	NNS	B-NP	O
.	.	.	O	O

–	–	LS	B-NP	O
Transform	Transform	NN	I-NP	B-protein
encoding	encode	VBG	B-VP	O
matrices	matrix	NNS	B-NP	O
to	to	TO	B-PP	O
tensors	tensor	NNS	B-NP	O
using	use	VBG	B-VP	O
K	K	NN	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
.	.	.	O	O

–	–	LS	B-LST	O
Initialize	Initialize	NNP	B-NP	O
α	α	NNP	I-NP	O
and	and	CC	O	O
β	β	VB	B-VP	O
randomly	randomly	RB	B-ADVP	O
.	.	.	O	O

–	–	SYM	O	O
t	t	NN	B-NP	O
=	=	SYM	B-VP	O
0	0	CD	B-NP	O
{	{	NN	I-NP	O
Iterative	Iterative	JJ	I-NP	O
step	step	NN	I-NP	O
}	}	.	O	O

t	t	NN	B-NP	O
←	←	NN	I-NP	O
t	t	NN	I-NP	O
+	+	SYM	B-NP	O
1	1	CD	I-NP	O

Compute	Compute	NN	B-NP	O
α	α	NN	I-NP	O
(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O
using	use	VBG	B-VP	O
equation	equation	NN	B-NP	O
(	(	(	O	O
15	15	CD	B-NP	O
)	)	)	O	O

Compute	Compute	NN	B-NP	O
β	β	NN	I-NP	O
(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O
using	use	VBG	B-VP	O
equation	equation	NN	B-NP	O
(	(	(	O	O
16	16	CD	B-NP	O
)	)	)	O	O


repeat	repeat	NN	B-NP	O


until	until	IN	B-SBAR	O
(	(	(	O	O
(	(	(	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	O	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O
−α	−α	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	O	O
)	)	)	O	O
α	α	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


≤	≤	LS	B-NP	O

1	
1	NN	I-NP	O
)	)	)	O	O
and	and	CC	O	O
(	(	(	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	O	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
t	t	NN	B-NP	O
)	)	)	O	O
−β	−β	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	O	O
)	)	)	O	O
β	β	NN	B-NP	O
(	(	(	O	O
t−1	t−1	NN	B-NP	O
)	)	)	O	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O
2	2	CD	B-NP	O


≤	≤	SYM	O	O

2	
2	CD	O	O
)	)	)	O	O
)	)	)	O	O
or	or	CC	O	O
(	(	(	O	O
t	t	NN	B-NP	O
>	>	SYM	O	O
tM	tM	NN	B-NP	O
ax	ax	NN	I-NP	O
)	)	)	O	O


3	3	CD	B-NP	O
Experimental	Experimental	NNP	I-NP	O
Evaluation	Evaluation	NNP	I-NP	O


This	This	DT	B-NP	O
section	section	NN	I-NP	O
presents	present	VBZ	B-VP	O
experimental	experimental	JJ	B-NP	O
evaluation	evaluation	NN	I-NP	O
in	in	IN	B-PP	O
comparing	compare	VBG	B-VP	O
the	the	DT	B-NP	O
proposed	propose	VBN	I-NP	O
method	method	NN	I-NP	O

TRM	TRM	NN	B-NP	B-protein
(	(	(	O	O
stands	stand	NNS	B-NP	O
for	for	IN	B-PP	O
‘tensor	‘tensor	NN	B-NP	O
regression	regression	NN	I-NP	O
model’	model’	NN	I-NP	O
)	)	)	O	O
with	with	IN	B-PP	O
the	the	DT	B-NP	O
most	most	RBS	I-NP	O
recent	recent	JJ	I-NP	O
reported	report	VBN	I-NP	O
meth-	meth-	NN	I-NP	O

ods	od	NNS	B-NP	O
for	for	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
prediction	prediction	NN	I-NP	O
on	on	IN	B-PP	O
commonly	commonly	RB	B-NP	O
used	use	VBN	I-NP	O
datasets.	datasets.	NNS	I-NP	O
Dis-	Dis-	.	O	O

cussion	cussion	NN	B-NP	O
on	on	IN	B-PP	O
the	the	DT	B-NP	O
framework	framework	NN	I-NP	O
and	and	CC	I-NP	O
methods	method	NNS	I-NP	O
will	will	MD	B-VP	O
follow	follow	VB	I-VP	O
the	the	DT	B-NP	O
experiment	experiment	NN	I-NP	O
report	report	NN	I-NP	O
.	.	.	O	O

Comparative	Comparative	JJ	B-NP	O
Evaluation.	Evaluation.	NNP	I-NP	O
The	The	DT	B-NP	O
comparison	comparison	NN	I-NP	O
is	be	VBZ	B-VP	O
carried	carry	VBN	I-VP	O
out	out	RP	B-PRT	O
using	use	VBG	B-VP	O
four	four	CD	B-NP	O
data	datum	NNS	I-NP	O
sets	set	NNS	I-NP	O


–	–	PDT	B-NP	O
The	The	DT	I-NP	O
Huesken	Huesken	NNP	I-NP	O
dataset	dataset	NN	I-NP	O
of	of	IN	B-PP	O
2431	2431	CD	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
targeting	target	VBG	B-VP	O
34	34	CD	B-NP	O
human	human	JJ	I-NP	O
and	and	CC	I-NP	O
ro-	ro-	JJ	I-NP	O

dent	dent	JJ	B-NP	B-RNA
mRNAs	mRNA	NNS	I-NP	I-RNA
,	,	,	O	O
commonly	commonly	RB	B-VP	O
divided	divide	VBN	I-VP	O
into	into	IN	B-PP	O
the	the	DT	B-NP	O
training	training	NN	I-NP	B-protein
set	set	NN	I-NP	I-protein
HU	HU	NN	I-NP	I-protein
train	train	NN	I-NP	I-protein
of	of	IN	B-PP	O
2182	2182	CD	B-NP	O
siR-	siR-	NN	I-NP	O

NAs	NA	NNS	B-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
testing	testing	NN	I-NP	O
set	set	NN	I-NP	O
HU	HU	NN	I-NP	O
test	test	NN	I-NP	O
of	of	IN	B-PP	O
249	249	CD	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
[	[	(	O	O
5	5	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O


–	–	PDT	B-NP	O
The	The	DT	I-NP	O
Reynolds	Reynolds	NNP	I-NP	O
dataset	dataset	NN	I-NP	O
of	of	IN	B-PP	O
240	240	CD	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
[	[	(	O	O
10	10	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O

–	–	PDT	B-NP	O
The	The	DT	I-NP	O
Vicker	Vicker	NNP	I-NP	O
dataset	dataset	NN	I-NP	O
of	of	IN	B-PP	O
76	76	CD	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
targeting	target	VBG	B-VP	O
two	two	CD	B-NP	O
genes	gene	NNS	I-NP	O
[	[	(	O	O
17	17	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O

–	–	PDT	B-NP	O
The	The	DT	I-NP	O
Harborth	Harborth	NNP	I-NP	O
dataset	dataset	NN	I-NP	O
of	of	IN	B-PP	O
44	44	CD	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
targeting	target	VBG	B-VP	O
one	one	CD	B-NP	O
gene	gene	NN	I-NP	O
[	[	(	O	O
3	3	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O


TRM	TRM	NN	B-NP	B-protein
is	be	VBZ	B-VP	O
compared	compare	VBN	I-VP	O
to	to	TO	B-PP	O
most	most	RBS	B-NP	O
state-of-the-art	state-of-the-art	JJ	I-NP	O
methods	method	NNS	I-NP	O
for	for	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	NN	I-NP	O

efficacy	efficacy	NN	B-NP	O
prediction	prediction	NN	I-NP	O
recently	recently	RB	B-VP	O
reported	report	VBN	I-VP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
literature.	literature.	NN	I-NP	O
As	As	IN	B-PP	O
experiments	experiment	NNS	B-NP	O
in	in	IN	B-PP	O
those	those	DT	B-NP	O

methods	method	NNS	B-NP	O
can	can	MD	B-VP	O
not	not	RB	I-VP	O
be	be	VB	I-VP	O
repeated	repeat	VBN	I-VP	O
directly	directly	RB	B-ADVP	O
,	,	,	O	O
we	we	PRP	B-NP	O
employed	employ	VBD	B-VP	O
the	the	DT	B-NP	O
results	result	NNS	I-NP	O
reported	report	VBN	B-VP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O

literature	literature	NN	B-NP	O
and	and	CC	O	O
carried	carry	VBD	B-VP	O
out	out	RP	B-PRT	O
experiments	experiment	NNS	B-NP	O
on	on	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
in	in	IN	B-PP	O
the	the	DT	B-NP	O
same	same	JJ	I-NP	O
conditions	condition	NNS	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O

other	other	JJ	B-NP	O
works.	works.	NNS	I-NP	O
Concretely	Concretely	RB	B-ADVP	O
,	,	,	O	O
the	the	DT	B-NP	O
comparative	comparative	JJ	I-NP	O
evaluation	evaluation	NN	I-NP	O
is	be	VBZ	B-VP	O
done	do	VBN	I-VP	O
as	as	IN	B-SBAR	O
follows	follow	VBZ	B-VP	O


1.	1.	LS	B-NP	O
Comparison	Comparison	NN	I-NP	O
of	of	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
with	with	IN	B-PP	O
Multiple	Multiple	NNP	B-NP	O
Kernel	Kernel	NNP	I-NP	O
Support	Support	NNP	I-NP	O
Vector	Vector	NNP	I-NP	O
Machine	Machine	NNP	I-NP	O
proposed	propose	VBD	B-VP	O

by	by	IN	B-PP	O
Qui	Qui	NNP	B-NP	O
et	et	FW	I-NP	O
al.	al.	FW	I-NP	O
[	[	(	O	O
8	8	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
The	The	DT	B-NP	O
author	author	NN	I-NP	O
of	of	IN	B-PP	O
[	[	(	O	O
8	8	CD	B-NP	O
]	]	)	O	O
reported	report	VBD	B-VP	O
their	their	PRP$	B-NP	O
Pearson	Pearson	NNP	I-NP	O
correlation	correlation	NN	I-NP	O
coeffi-	coeffi-	NN	I-NP	O

cient	cient	JJ	O	O
(	(	(	O	O
R	R	NN	B-NP	O
)	)	)	O	O
of	of	IN	B-PP	O
0.62	0.62	CD	B-NP	O
obtained	obtain	VBN	B-VP	O
by	by	IN	B-PP	O
10-fold	10-fold	JJ	B-NP	O
cross	cross	NN	I-NP	O
validation	validation	NN	I-NP	O
on	on	IN	B-PP	O
the	the	DT	B-NP	O
whole	whole	JJ	I-NP	O
Huesken	Huesken	NNP	I-NP	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


409	409	CD	B-NP	O


Table	Table	JJ	B-ADJP	O
2.	2.	LS	B-LST	O
The	The	DT	B-NP	O
R	R	NN	I-NP	O
values	value	NNS	I-NP	O
of	of	IN	B-PP	O
18	18	CD	B-NP	O
models	model	NNS	I-NP	O
and	and	CC	O	O
TRM	TRM	NN	B-NP	B-protein
on	on	IN	B-PP	O
three	three	CD	B-NP	O
independent	independent	JJ	I-NP	O
data	datum	NNS	I-NP	O
sets	set	NNS	I-NP	O


Algorithm	Algorithm	NNP	B-NP	O


RReynolds	RReynold	NNS	B-NP	O

(	(	(	B-LST	O
244si/7g	244si/7g	LS	I-LST	O
)	)	)	O	O

GPboot	GPboot	NN	B-NP	O

0.55	0.55	CD	B-NP	O

Uitei	Uitei	NNP	B-NP	O

0.47	0.47	CD	B-NP	O

Amarzguioui	Amarzguioui	IN	B-PP	O
0.45	0.45	CD	B-NP	O

0.03	0.03	CD	B-NP	O

Hsieh	Hsieh	NN	B-NP	O

0.03	0.03	CD	B-NP	O

Takasaki	Takasaki	NNP	B-NP	O

Reynolds	Reynold	NNS	B-NP	O
1	1	CD	I-NP	O

0.35	0.35	CD	B-NP	O

0.37	0.37	CD	B-NP	O

Reynolds	Reynold	NNS	B-NP	O
2	2	CD	B-NP	O

0.29	0.29	CD	B-NP	O

Schawarz	Schawarz	NNP	B-NP	O

Khvorova	Khvorova	NNP	B-NP	O

0.15	0.15	CD	B-NP	O


RV	RV	NN	B-NP	O
icker	icker	NN	I-NP	O

(	(	(	B-LST	O
76si/2g	76si/2g	LS	I-LST	O
)	)	)	O	O

0.35	0.35	CD	B-NP	O

0.58	0.58	CD	B-NP	O

0.47	0.47	CD	B-NP	O

0.15	0.15	CD	B-NP	O

0.25	0.25	CD	B-NP	O

0.47	0.47	CD	B-NP	O

0.44	0.44	CD	B-NP	O

0.35	0.35	CD	B-NP	O

0.19	0.19	CD	B-NP	O


RHarborth	RHarborth	NN	B-NP	O

(	(	(	B-LST	O
44si/1g	44si/1g	LS	I-LST	O
)	)	)	O	O

0.43	0.43	CD	B-NP	O

0.31	0.31	CD	B-NP	O

0.34	0.34	CD	B-NP	O

0.17	0.17	CD	B-NP	O

0.01	0.01	CD	B-NP	O

0.23	0.23	CD	B-NP	O

0.23	0.23	CD	B-NP	O

0.01	0.01	CD	B-NP	O

0.11	0.11	CD	B-NP	O


Algorithm	Algorithm	NNP	B-NP	O


RReynolds	RReynold	NNS	B-NP	O

(	(	(	B-LST	O
244si/7g	244si/7g	LS	I-LST	O
)	)	)	O	O


Stockholm	Stockholm	NNP	B-NP	O
1	1	CD	I-NP	O
0.05	0.05	CD	I-NP	O

Stockholm	Stockholm	NNP	B-NP	O
2	2	CD	I-NP	O
0.00	0.00	CD	I-NP	O

0.11	0.11	CD	B-NP	O

Tree	Tree	NNP	B-NP	O

0.33	0.33	CD	B-NP	O

Luo	Luo	NNP	B-NP	O

0.54	0.54	CD	B-NP	O

i-score	i-score	NN	B-NP	O

Biopredsi	Biopredsi	NNP	B-NP	O

0.53	0.53	CD	B-NP	O

0.54	0.54	CD	B-NP	O

DSIR	DSIR	NN	B-NP	B-protein

0.40	0.40	CD	B-NP	O

Katoh	Katoh	NNP	B-NP	O

0.54	0.54	CD	B-NP	O

SVM	SVM	NN	B-NP	O

TRM	TRM	NN	B-NP	B-protein

0.60	0.60	CD	B-NP	O


RV	RV	NN	B-NP	O
icker	icker	NN	I-NP	O

(	(	(	B-LST	O
76si/2g	76si/2g	LS	I-LST	O
)	)	)	O	O

0.18	0.18	CD	B-NP	O

0.15	0.15	CD	B-NP	O

0.43	0.43	CD	B-NP	O

0.27	0.27	CD	B-NP	O

0.58	0.58	CD	B-NP	O

0.57	0.57	CD	B-NP	O

0.49	0.49	CD	B-NP	O

0.43	0.43	CD	B-NP	O

0.52	0.52	CD	B-NP	O

0.58	0.58	CD	B-NP	O


RHarborth	RHarborth	NN	B-NP	O

(	(	(	B-LST	O
44si/1g	44si/1g	LS	I-LST	O
)	)	)	O	O

0.28	0.28	CD	B-NP	O

0.41	0.41	CD	B-NP	O

0.06	0.06	CD	B-NP	O

0.40	0.40	CD	B-NP	O

0.43	0.43	CD	B-NP	O

0.51	0.51	CD	B-NP	O

0.51	0.51	CD	B-NP	O

0.44	0.44	CD	B-NP	O

0.54	0.54	CD	B-NP	O

0.55	0.55	CD	B-NP	O


dataset.	dataset.	LS	B-LST	O
The	The	DT	B-NP	O
Pearson	Pearson	NNP	I-NP	O
correlation	correlation	NN	I-NP	O
coefficient	coefficient	NN	I-NP	O
(	(	(	O	O
R	R	NN	B-NP	B-protein
)	)	)	O	O
is	be	VBZ	B-VP	O
carefully	carefully	RB	I-VP	O
evaluated	evaluate	VBN	I-VP	O
by	by	IN	B-PP	O

TRM	TRM	NN	B-NP	B-protein
by	by	IN	B-PP	O
10	10	CD	B-NP	O
times	time	NNS	I-NP	O
of	of	IN	B-PP	O
10-fold	10-fold	JJ	B-NP	O
cross	cross	NN	I-NP	O
validation	validation	NN	I-NP	O
with	with	IN	B-PP	O
the	the	DT	B-NP	O
average	average	JJ	I-NP	O
value	value	NN	I-NP	O
of	of	IN	B-PP	O
0.64	0.64	CD	B-NP	O
.	.	.	O	O


2.	2.	LS	B-NP	O
Comparison	Comparison	NN	I-NP	O
of	of	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
with	with	IN	B-PP	O
four	four	CD	B-NP	O
state-of-the-art	state-of-the-art	JJ	I-NP	O
methods	method	NNS	I-NP	O
of	of	IN	B-PP	O
BIOPREDsi	BIOPREDsi	NNP	B-NP	O
[	[	(	O	O
5	5	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O

DSIR	DSIR	NN	B-NP	B-protein
[	[	(	O	O
16	16	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
Thermocomposition21	Thermocomposition21	NN	B-NP	B-protein
[	[	(	O	O
13	13	CD	B-NP	O
]	]	)	O	O
,	,	,	O	O
SVM	SVM	NN	B-NP	O
[	[	(	O	O
12	12	CD	B-NP	O
]	]	)	O	O
by	by	IN	B-PP	O
HU	HU	NN	B-NP	O
train	train	NN	I-NP	O
and	and	CC	O	O
HU	HU	NN	B-NP	O
test	test	NN	I-NP	O
.	.	.	O	O

The	The	DT	B-NP	O
Pearson	Pearson	NNP	I-NP	O
correlation	correlation	NN	I-NP	O
coefficients	coefficient	NNS	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
four	four	CD	I-NP	O
models	model	NNS	I-NP	O
BIOPREDsi	BIOPREDsi	NN	I-NP	O
,	,	,	O	O
DSIR	DSIR	NN	B-NP	B-protein
,	,	,	O	O

Thermocomposition21	Thermocomposition21	NN	B-NP	B-protein
and	and	CC	I-NP	O
SVM	SVM	NN	I-NP	O
are	be	VBP	B-VP	O
0.66	0.66	CD	B-NP	O
,	,	,	I-NP	O
0.67	0.67	CD	I-NP	O
,	,	,	I-NP	O
0.66	0.66	CD	I-NP	O
and	and	CC	I-NP	O
0.80	0.80	CD	I-NP	O
,	,	,	O	O
respectively	respectively	RB	B-ADVP	O
.	.	.	O	O

The	The	DT	B-NP	O
performance	performance	NN	I-NP	O
of	of	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
estimated	estimate	VBN	B-VP	O
on	on	IN	B-PP	O
HU	HU	NN	B-NP	O
test	test	NN	I-NP	O
is	be	VBZ	B-VP	O
0.68	0.68	CD	B-NP	O
that	that	WDT	B-NP	O
is	be	VBZ	B-VP	O
slightly	slightly	RB	B-ADJP	O
higher	high	JJR	I-ADJP	O

than	than	IN	B-PP	O
that	that	DT	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
first	first	JJS	I-NP	O
three	three	CD	I-NP	O
models	model	NNS	I-NP	O
but	but	CC	O	O
lower	low	JJR	B-ADJP	O
than	than	IN	B-PP	O
that	that	DT	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
last	last	JJ	I-NP	O
model	model	NN	I-NP	O
.	.	.	O	O


3.	3.	LS	B-NP	O
Comparison	Comparison	NN	I-NP	O
of	of	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
with	with	IN	B-PP	O
18	18	CD	B-NP	O
methods	method	NNS	I-NP	O
including	include	VBG	B-PP	O
BIOPREDsi	BIOPREDsi	NNP	B-NP	O
,	,	,	O	O
DSIR	DSIR	NNP	B-NP	O
,	,	,	O	O
Ther-	Ther-	NNP	B-NP	O

mocomposition21	mocomposition21	NN	B-NP	O
,	,	,	O	O
SVM	SVM	NN	B-NP	O
when	when	WRB	B-ADVP	O
training	train	VBG	B-VP	O
on	on	IN	B-PP	O
HU	HU	NN	B-NP	O
train	train	NN	I-NP	O
and	and	CC	I-NP	O
testing	testing	NN	I-NP	O
on	on	IN	B-PP	O
three	three	CD	B-NP	O

independent	independent	JJ	B-NP	O
datasets	dataset	NNS	I-NP	O
of	of	IN	B-PP	O
Reynolds	Reynolds	NNP	B-NP	O
,	,	,	I-NP	O
Vicker	Vicker	NNP	I-NP	O
and	and	CC	I-NP	O
Harborth	Harborth	NNP	I-NP	O
as	as	IN	B-SBAR	O
reported	report	VBN	B-VP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O

recent	recent	JJ	B-NP	O
article	article	NN	I-NP	O
[	[	(	O	O
12	12	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
As	As	IN	B-SBAR	O
shown	show	VBN	B-VP	O
in	in	IN	B-PP	O
Table	Table	JJ	B-NP	O
2	2	CD	I-NP	O
(	(	(	O	O
taken	take	VBN	B-VP	O
from	from	IN	B-PP	O
[	[	(	O	O
12	12	CD	B-NP	O
]	]	)	O	O
with	with	IN	B-PP	O
the	the	DT	B-NP	O
added	add	VBN	I-NP	O
last	last	JJ	B-ADJP	O

row	row	NN	B-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
TRM	TRM	NN	I-NP	O
result	result	NN	I-NP	O
)	)	)	O	O
,	,	,	O	O
TRM	TRM	NN	B-NP	B-protein
considerably	considerably	RB	B-ADVP	O
achieved	achieve	VBD	B-VP	O
results	result	NNS	B-NP	O
higher	high	JJR	B-ADJP	O
than	than	IN	B-PP	O
all	all	DT	B-NP	O

of	of	IN	B-PP	O
18	18	CD	B-NP	O
methods	method	NNS	I-NP	O
on	on	IN	B-PP	O
the	the	DT	B-NP	O
all	all	DT	I-NP	O
three	three	CD	I-NP	O
independent	independent	JJ	I-NP	O
testing	testing	NN	I-NP	O
datasets	dataset	NNS	I-NP	O
.	.	.	O	O


In	In	IN	B-PP	O
running	run	VBG	B-VP	O
Algorithm	Algorithm	NN	B-NP	O
2	2	CD	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
thresholds	threshold	NNS	I-NP	O
for	for	IN	B-PP	O
the	the	DT	B-NP	O
weight	weight	NN	I-NP	O
vector	vector	NN	I-NP	O
α	α	NN	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
coef-	coef-	NN	I-NP	O

ficient	ficient	JJ	B-NP	O
vector	vector	NN	I-NP	O
β	β	NN	I-NP	O
are	be	VBP	B-VP	O
set	set	VBN	I-VP	O
up	up	RP	B-PRT	O
as	as	IN	B-PP	O
0.001	0.001	CD	B-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
maximum	maximum	JJ	I-NP	O
iterative	iterative	JJ	I-NP	O
step	step	NN	I-NP	O
is	be	VBZ	B-VP	O
1000.	1000.	CD	B-NP	O
The	The	DT	O	O

turning	turn	VBG	B-VP	O
parameter	parameter	NN	B-NP	O
λ	λ	NN	I-NP	O
is	be	VBZ	B-VP	O
chosen	choose	VBN	I-VP	O
by	by	IN	B-PP	O
minimizing	minimize	VBG	B-VP	O
the	the	DT	B-NP	O
risk	risk	NN	I-NP	O
function	function	NN	I-NP	O
when	when	WRB	B-ADVP	O
testing	test	VBG	B-VP	O
on	on	IN	B-PP	O

validation	validation	NN	B-NP	O
dataset.	dataset.	NN	I-NP	O
Particularly	Particularly	RB	B-ADVP	O
,	,	,	O	O
we	we	PRP	B-NP	O
do	do	VBP	B-VP	O
10–fold	10–fold	RB	I-VP	O
cross	cross	VB	I-VP	O
validation	validation	NN	B-NP	O
on	on	IN	B-PP	O
the	the	DT	B-NP	O
training	training	NN	I-NP	O

set	set	VBN	B-VP	O
for	for	IN	B-PP	O
each	each	DT	B-NP	O
λ	λ	NN	I-NP	O
belonging	belong	VBG	B-VP	O
to	to	TO	B-PP	O
[	[	(	B-NP	O
0	0	CD	I-NP	O
,	,	,	O	O
log	log	NN	B-NP	O
50	50	CD	I-NP	O
]	]	)	O	O
and	and	CC	O	O
compute	compute	VB	B-VP	O
the	the	DT	B-NP	O
risk	risk	NN	I-NP	O
function	function	NN	I-NP	O


⎛	⎛	LS	B-LST	O


⎞	⎞	LS	B-LST	O


R	R	NN	B-NP	B-protein
(	(	(	O	O
λ	λ	NN	B-NP	O
)	)	)	O	O
=	=	.	O	O


1	1	CD	B-NP	O

F	F	NN	B-NP	O


F	F	NN	B-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
3	3	LS	B-LST	O
)	)	)	O	O


i=1	i=1	NN	B-NP	O


⎝	⎝	LS	B-LST	O


1	1	CD	B-NP	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
11	11	CD	O	O
)	)	)	O	O
f	f	NN	B-NP	O
oldi	oldi	NN	I-NP	O
(	(	(	O	O
cid	cid	NN	B-NP	O
:	:	:	O	O
11	11	CD	B-NP	O
)	)	)	O	O


(	(	(	O	O
cid	cid	JJ	B-ADJP	O
:	:	:	O	O
3	3	CD	B-NP	O
)	)	)	O	O


xj∈f	xj∈f	NN	B-NP	O
oldi	oldi	NNS	I-NP	O


(	(	(	O	O
yj	yj	NN	B-NP	O
−	−	NN	I-NP	O
f	f	NN	I-NP	O
(	(	(	O	O
xj	xj	NN	B-NP	O
)	)	)	O	O
)	)	)	O	O


2	2	LS	B-LST	O


⎠	⎠	LS	B-LST	O


where	where	WRB	B-ADVP	O
f	f	NN	B-NP	O
oldi	oldi	NN	I-NP	O
is	be	VBZ	B-VP	O
validation	validation	NN	B-NP	O
set	set	NN	I-NP	O
,	,	,	O	O
f	f	NN	B-NP	O
(	(	(	O	O
x	x	NN	B-NP	O
)	)	)	O	O
is	be	VBZ	B-VP	O
a	a	DT	B-NP	O
tensor	tensor	NN	I-NP	O
predictor	predictor	NN	I-NP	O
learnt	learnt	NN	I-NP	O
from	from	IN	B-PP	O
training	training	NN	B-NP	O
set	set	NN	I-NP	O

except	except	IN	B-PP	O
validation	validation	NN	B-NP	B-protein
set	set	NN	I-NP	I-protein
f	f	NN	I-NP	I-protein
oldi.	oldi.	NN	I-NP	I-protein
F	F	NN	I-NP	I-protein
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
number	number	NN	I-NP	O
of	of	IN	B-PP	O
folds	fold	NNS	B-NP	O
to	to	TO	B-VP	O
do	do	VB	I-VP	O
cross	cross	VB	I-VP	O
validation	validation	NN	B-NP	O
on	on	IN	B-PP	O

training	training	NN	B-NP	O
set.	set.	NN	I-NP	O
In	In	IN	B-PP	O
our	our	PRP$	B-NP	O
work	work	NN	I-NP	O
,	,	,	O	O
we	we	PRP	B-NP	O
do	do	VBP	B-VP	O
F-fold	F-fold	JJ	B-NP	O
cross	cross	NN	I-NP	O
validation	validation	NN	I-NP	O
thus	thus	RB	O	O
F	F	NN	B-NP	O
equals	equal	NNS	I-NP	O
to	to	TO	B-PP	O
10	10	CD	B-NP	O
.	.	.	O	O


In	In	IN	B-PP	O
the	the	DT	B-NP	O
transformation	transformation	NN	I-NP	B-protein
matrices	matrix	NNS	I-NP	I-protein
learning	learn	VBG	B-VP	I-protein
task	task	NN	B-NP	I-protein
,	,	,	O	O
we	we	PRP	B-NP	O
use	use	VBP	B-VP	O
the	the	DT	B-NP	O
labelled	label	VBN	I-NP	O
dataset	dataset	NN	I-NP	O
col-	col-	NN	I-NP	O

lected	lecte	VBN	B-VP	O
from	from	IN	B-PP	O
siRecord	siRecord	NN	B-NP	O
database	database	NN	I-NP	O
[	[	(	O	O
9	9	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O
This	This	DT	B-NP	O
data	datum	NNS	I-NP	O
set	set	NN	I-NP	O
has	have	VBZ	B-VP	O
2470	2470	CD	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
in	in	IN	B-PP	O

‘very	‘very	JJ	B-NP	O
high’	high’	NN	I-NP	O
class	class	NN	I-NP	O
and	and	CC	O	O
2514	2514	CD	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA
in	in	IN	B-PP	O
‘low’	‘low’	NN	B-NP	O
and	and	CC	I-NP	O
‘medium’	‘medium’	NN	I-NP	O
classes.	classes.	NNS	I-NP	O
Each	Each	NNP	B-NP	O

siRNA	siRNA	NN	B-NP	B-DNA
sequence	sequence	NN	I-NP	I-DNA
has	have	VBZ	B-VP	O
19	19	CD	B-NP	O
nucleotides.	nucleotides.	NN	I-NP	O
Seven	Seven	CD	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
used	use	VBN	B-VP	O
to	to	TO	B-VP	O
learn	learn	VB	I-VP	O
matrices	matrix	NNS	B-NP	O

are	be	VBP	O	O
Reynolds	Reynolds	NNP	B-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
Uitei	Uitei	NNP	B-NP	O
Rule	Rule	NNP	I-NP	O
,	,	,	O	O
Amarzguioui	Amarzguioui	NNP	B-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
Jalag	Jalag	NN	B-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
Hsieh	Hsieh	NN	B-NP	O
rule	rule	NN	I-NP	O
,	,	,	O	O
Takasaki	Takasaki	NNP	B-NP	O


410	410	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


Table	Table	JJ	B-ADJP	O
3.	3.	LS	B-LST	O
The	The	DT	B-NP	O
learnt	learnt	NN	I-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
containing	contain	VBG	B-VP	O
characteristics	characteristic	NNS	B-NP	O
of	of	IN	B-PP	O
Reynolds	Reynold	NNS	B-NP	O
rule	rule	NN	I-NP	O


1	1	CD	B-NP	O


2	2	LS	B-LST	O


3	3	LS	B-LST	O


4	4	LS	B-LST	O


10	10	CD	B-NP	O


11	11	CD	B-NP	O


12	12	CD	B-NP	O


19	19	CD	B-NP	O


A	A	DT	B-NP	O
0.29704	0.29704	CD	I-NP	O
0.217977	0.217977	CD	I-NP	O
0.423469	0.423469	CD	I-NP	O
0.266597	0.266597	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
0.363636	0.363636	CD	B-NP	O
0.246021	0.246021	CD	I-NP	O
0.224727	0.224727	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	.	O	O
0.393939	0.393939	CD	B-NP	O

C	C	NN	B-NP	O
0.231159	0.231159	CD	I-NP	O
0.235744	0.235744	CD	I-NP	O
0.255102	0.255102	CD	I-NP	O
0.226922	0.226922	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	.	O	O

0.252513	0.252513	CD	B-NP	O
0.267744	0.267744	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	.	O	O
0.0757576	0.0757576	CD	B-NP	O

G	G	VBG	B-VP	O
0.155341	0.155341	CD	B-NP	O
0.211418	0.211418	CD	I-NP	O
0.0459184	0.0459184	CD	I-NP	O
0.237968	0.237968	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
0.229437	0.229437	CD	B-NP	O
0.221336	0.221336	CD	I-NP	O
0.260756	0.260756	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	.	O	O
0.161616	0.161616	CD	B-NP	O

0.27551	0.27551	CD	B-NP	O
0.268513	0.268513	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	''	O	O
0.406926	0.406926	CD	B-NP	O
0.28013	0.28013	CD	I-NP	O
0.246773	0.246773	CD	I-NP	O
.	.	.	O	O
.	.	.	O	O
.	.	.	O	O
0.368687	0.368687	CD	B-NP	O

U	U	NN	B-NP	O
0.31646	0.31646	CD	I-NP	O
0.33486	0.33486	CD	I-NP	O


0	0	CD	B-NP	O


rule	rule	NN	B-NP	O
and	and	CC	O	O
Huesken	Huesken	NNP	B-NP	O
rule.	rule.	VBP	B-VP	O
The	The	DT	B-NP	O
convergence	convergence	NN	I-NP	O
criteria	criterion	NNS	I-NP	O
in	in	IN	B-PP	O
Algorithm	Algorithm	NNP	B-NP	O
1	1	CD	I-NP	O
are	be	VBP	B-VP	O
set	set	VBN	I-VP	O
up	up	RP	B-PRT	O
as	as	IN	B-SBAR	O

−8	−8	CD	B-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
maximum	maximum	NN	I-NP	O

following	follow	VBG	B-PP	O
:	:	:	O	O
threshold	threshold	NN	B-NP	O

	
	NN	I-NP	O
for	for	IN	B-PP	O
transformation	transformation	NN	B-NP	O
matrices	matrix	NNS	I-NP	O
is	be	VBZ	B-VP	O
2.5E	2.5E	NN	B-NP	O

iterative	iterative	JJ	B-NP	O
step	step	NN	I-NP	O
is	be	VBZ	B-VP	O
5000	5000	CD	B-NP	O
.	.	.	O	O


Discussion.	Discussion.	NN	B-NP	O
As	As	IN	B-SBAR	O
reported	report	VBN	B-VP	O
in	in	IN	B-PP	O
the	the	DT	B-NP	O
experimental	experimental	JJ	I-NP	O
comparative	comparative	JJ	I-NP	O
evaluation	evaluation	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
pro-	pro-	JJ	I-NP	O

posed	pose	VBN	B-NP	O
TRM	TRM	NN	I-NP	B-protein
achieved	achieve	VBD	B-VP	O
higher	high	JJR	B-NP	O
results	result	NNS	I-NP	O
than	than	IN	B-PP	O
most	most	JJS	B-NP	O
other	other	JJ	I-NP	O
methods	method	NNS	I-NP	O
for	for	IN	B-PP	O
prediction	prediction	NN	B-NP	O
of	of	IN	B-PP	O

siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	FW	B-ADVP	O
efficacy.	efficacy.	FW	I-ADVP	O
There	There	EX	B-NP	O
are	be	VBP	B-VP	O
some	some	DT	B-NP	O
reasons	reason	NNS	I-NP	O
of	of	IN	B-PP	O
that.	that.	IN	B-PP	O
First	First	NNP	B-NP	O
,	,	,	O	O
it	it	PRP	B-NP	O
is	be	VBZ	B-VP	O
expensive	expensive	JJ	B-ADJP	O

and	and	CC	O	O
hard	hard	JJ	B-ADJP	O
to	to	TO	B-VP	O
analyze	analyze	VB	I-VP	O
the	the	DT	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
of	of	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	B-RNA
,	,	,	O	O
and	and	CC	O	O
thus	thus	RB	O	O
most	most	RBS	B-ADJP	O
available	available	JJ	I-ADJP	O

datasets	dataset	NNS	B-NP	O
are	be	VBP	B-VP	O
of	of	IN	B-PP	O
relatively	relatively	RB	B-NP	O
small	small	JJ	I-NP	O
size	size	NN	I-NP	O
leading	lead	VBG	B-VP	O
to	to	TO	B-PP	O
limited	limited	JJ	B-NP	O
results.	results.	NNS	I-NP	O
Second	Second	NNP	B-NP	O
,	,	,	O	O
TRM	TRM	NNP	B-NP	O
has	have	VBZ	B-VP	O

its	its	PRP$	B-NP	O
advantages	advantage	NNS	I-NP	O
by	by	IN	B-PP	O
incorporating	incorporate	VBG	B-VP	O
domain	domain	NN	B-NP	O
knowledge	knowledge	NN	I-NP	O
(	(	(	O	O
siRNA	siRNA	NN	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
)	)	)	O	O
found	find	VBD	B-VP	O

from	from	IN	B-PP	O
different	different	JJ	B-NP	O
datasets	dataset	NNS	I-NP	O
in	in	IN	B-PP	O
experiments.	experiments.	JJ	B-NP	O
Third	Third	NNP	I-NP	O
,	,	,	O	O
TRM	TRM	NNP	B-NP	O
is	be	VBZ	B-VP	O
generic	generic	JJ	B-ADJP	O
and	and	CC	O	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
eas-	eas-	JJ	B-ADJP	O

ily	ily	RB	B-ADVP	O
exploited	exploit	VBN	B-VP	O
when	when	WRB	B-ADVP	O
new	new	JJ	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
are	be	VBP	B-VP	O
discovered	discover	VBN	I-VP	O
or	or	CC	B-NP	O
more	more	RBR	I-NP	O
analyzed	analyze	VBN	I-NP	O
siRNAs	siRNA	NNS	I-NP	O

be	be	VB	B-VP	O
obtained.	obtained.	VBN	I-VP	O
Four	Four	CD	B-NP	O
,	,	,	O	O
one	one	CD	B-NP	O
drawback	drawback	NN	I-NP	O
of	of	IN	B-PP	O
TRM	TRM	NN	B-NP	B-protein
is	be	VBZ	B-VP	O
its	its	PRP$	B-NP	O
transformation	transformation	NN	I-NP	O
matrices	matrix	NNS	I-NP	O
are	be	VBP	B-VP	O

learned	learn	VBN	B-VP	O
using	use	VBG	B-VP	O
positional	positional	JJ	B-NP	O
features	feature	NNS	I-NP	O
of	of	IN	B-PP	O
available	available	JJ	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
,	,	,	O	O
and	and	CC	O	O
thus	thus	RB	O	O
they	they	PRP	B-NP	O
lack	lack	VBP	B-VP	O

some	some	DT	B-NP	O
characteristics	characteristic	NNS	I-NP	O
effecting	effecte	VBG	B-VP	O
to	to	TO	B-PP	O
knockdown	knockdown	JJ	B-NP	O
efficacy	efficacy	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
such	such	JJ	B-ADJP	O

as	as	IN	B-PP	O
GC	GC	NN	B-NP	O
content	content	NN	I-NP	O
,	,	,	O	O
thermodynamic	thermodynamic	JJ	B-NP	O
properties	property	NNS	I-NP	O
,	,	,	O	O
GC	GC	NN	B-NP	O
stretch	stretch	NN	I-NP	O
,	,	,	O	O
etc.	etc.	FW	B-NP	O
It	It	PRP	B-NP	O
may	may	MD	B-VP	O
be	be	VB	I-VP	O
one	one	CD	B-NP	O
of	of	IN	B-PP	O

reasons	reason	NNS	B-NP	O
that	that	IN	B-NP	O
at	at	IN	B-PP	O
this	this	DT	B-NP	O
moment	moment	NN	I-NP	O
TRM	TRM	NN	I-NP	O
can	can	MD	B-VP	O
not	not	RB	I-VP	O
get	get	VB	I-VP	O
higher	high	JJR	B-NP	O
performance	performance	NN	I-NP	O
when	when	WRB	B-ADVP	O
testing	test	VBG	B-VP	O

on	on	IN	B-PP	O
HU	HU	NN	B-NP	O
test	test	NN	I-NP	O
set	set	VBN	B-VP	O
than	than	IN	B-PP	O
the	the	DT	B-NP	O
best	good	JJS	I-NP	O
current	current	JJ	I-NP	O
model	model	NN	I-NP	O
SVM	SVM	NN	I-NP	O
[	[	(	O	O
12	12	CD	B-NP	O
]	]	)	O	O
.	.	.	O	O


Table	Table	JJ	B-NP	B-protein
3	3	CD	I-NP	I-protein
shows	show	VBZ	B-VP	O
the	the	DT	B-NP	O
learned	learn	VBN	I-NP	O
transformation	transformation	NN	I-NP	O
matrix	matrix	NN	I-NP	O
capturing	capture	VBG	B-VP	O
positional	positional	JJ	B-NP	O
charac-	charac-	NN	I-NP	O

teristics	teristic	NNS	B-NP	O
of	of	IN	B-PP	O
Reynolds	Reynolds	NNP	B-NP	O
rule.	rule.	NN	I-NP	O
One	One	CD	I-NP	O
of	of	IN	B-PP	O
characteristics	characteristic	NNS	B-NP	O
is	be	VBZ	B-VP	O
described	describe	VBN	I-VP	O
as	as	IN	B-PP	O
“An	“An	NN	B-NP	O
nucleotide	nucleotide	NN	I-NP	O

‘A’	‘A’	NN	B-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-DNA
19”.	19”.	CD	I-NP	I-DNA
That	That	IN	B-NP	O
characteristic	characteristic	JJ	B-NP	O
means	mean	NNS	I-NP	O
that	that	IN	B-NP	O
at	at	IN	B-PP	O
column	column	NN	B-NP	O
19	19	CD	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
cell	cell	NN	I-NP	O
(	(	(	O	O
1	1	CD	B-NP	O
,	,	,	I-NP	O
19	19	CD	I-NP	O
)	)	)	O	O

has	have	VBZ	B-VP	O
to	to	TO	I-VP	O
be	be	VB	I-VP	O
the	the	DT	B-NP	O
maximum	maximum	JJ	I-NP	O
value.	value.	NN	I-NP	O
In	In	IN	B-PP	O
the	the	DT	B-NP	O
matrix	matrix	NN	I-NP	O
,	,	,	O	O
the	the	DT	B-NP	O
value	value	NN	I-NP	O
at	at	IN	B-PP	O
this	this	DT	B-NP	O
cell	cell	NN	I-NP	O
is	be	VBZ	B-VP	O
0.393939	0.393939	CD	B-NP	O

and	and	CC	O	O
is	be	VBZ	B-VP	O
the	the	DT	B-NP	O
highest	high	JJS	I-NP	O
value	value	NN	I-NP	O
of	of	IN	B-PP	O
this	this	DT	B-NP	O
column.	column.	NN	I-NP	O
In	In	IN	B-PP	O
this	this	DT	B-NP	O
column	column	NN	I-NP	O
,	,	,	O	O
we	we	PRP	B-NP	O
also	also	RB	B-ADVP	O
know	know	VBP	B-VP	O
knock-	knock-	JJ	B-ADJP	O

down	down	RB	B-ADVP	O
efficacy	efficacy	NN	B-NP	O
of	of	IN	B-PP	O
each	each	DT	B-NP	O
nucleotide	nucleotide	NN	I-NP	O
at	at	IN	B-PP	O
position	position	NN	B-NP	B-protein
19.	19.	CD	I-NP	I-protein
Therefore	Therefore	RB	B-ADVP	O
,	,	,	O	O
nucleotides	nucleotide	NNS	B-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O

arranged	arrange	VBN	B-VP	O
by	by	IN	B-PP	O
the	the	DT	B-NP	O
decreasing	decrease	VBG	I-NP	O
order	order	NN	I-NP	O
of	of	IN	B-PP	O
their	their	PRP$	B-NP	O
efficacy	efficacy	NN	I-NP	O
:	:	:	O	O
A	A	DT	B-NP	O
,	,	,	O	O
U	U	NN	B-NP	O
,	,	,	O	O
G	G	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O
C.	C.	NNP	B-NP	O
In	In	NNP	I-NP	O
the	the	DT	B-NP	O
order	order	NN	I-NP	O
,	,	,	O	O

nucleotide	nucleotide	NN	B-NP	O
U	U	NN	I-NP	O
has	have	VBZ	B-VP	O
efficacy	efficacy	NN	B-NP	O
of	of	IN	B-PP	O
0.368687	0.368687	CD	B-NP	O
that	that	WDT	B-NP	O
also	also	RB	B-ADVP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
used	use	VBN	I-VP	O
to	to	TO	B-VP	O
design	design	VB	I-VP	O
effective	effective	JJ	B-NP	O

siRNAs.	siRNAs.	NN	B-NP	O
In	In	IN	B-PP	O
addition	addition	NN	B-NP	O
,	,	,	O	O
if	if	IN	B-SBAR	O
a	a	DT	B-NP	O
position	position	NN	I-NP	O
on	on	IN	B-PP	O
siRNAs	siRNA	NNS	B-NP	O
is	be	VBZ	B-VP	O
not	not	RB	I-VP	O
described	describe	VBN	I-VP	O
in	in	IN	B-PP	O
characteristics	characteristic	NNS	B-NP	O

of	of	IN	B-PP	O
the	the	DT	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
,	,	,	O	O
values	value	NNS	B-NP	O
at	at	IN	B-PP	O
the	the	DT	B-NP	O
column	column	NN	I-NP	O
corresponding	correspond	VBG	B-VP	O
to	to	TO	B-PP	O
this	this	DT	B-NP	O
position	position	NN	I-NP	O
is	be	VBZ	B-VP	O
learned	learn	VBN	I-VP	O

to	to	TO	B-VP	O
satisfy	satisfy	VB	I-VP	O
classification	classification	NN	B-NP	O
assumption	assumption	NN	I-NP	O
and	and	CC	I-NP	O
property	property	NN	I-NP	O
to	to	TO	B-VP	O
get	get	VB	I-VP	O
knockdown	knockdown	JJ	B-NP	O
efficacy	efficacy	NN	I-NP	O
of	of	IN	B-PP	O

each	each	DT	B-NP	O
nucleotide	nucleotide	NN	I-NP	O
such	such	JJ	B-PP	O
as	as	IN	I-PP	O
values	value	NNS	B-NP	O
at	at	IN	B-PP	O
columns	column	NNS	B-NP	O
1	1	CD	I-NP	O
,	,	,	I-NP	O
2	2	CD	I-NP	O
,	,	,	I-NP	O
4	4	CD	I-NP	O
and	and	CC	O	O
so	so	RB	B-ADVP	O
on	on	RB	I-ADVP	O
.	.	.	O	O


4	4	CD	B-NP	O
Conclusion	Conclusion	NN	I-NP	O


In	In	IN	B-PP	O
this	this	DT	B-NP	O
paper	paper	NN	I-NP	O
,	,	,	O	O
we	we	PRP	B-NP	O
have	have	VBP	B-VP	O
proposed	propose	VBN	I-VP	O
a	a	DT	B-NP	O
novel	novel	JJ	I-NP	O
framework	framework	NN	I-NP	O
to	to	TO	B-VP	O
predict	predict	VB	I-VP	O
knockdown	knockdown	RP	B-PRT	O
effica-	effica-	NN	B-NP	O

cies	cy	NNS	B-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
by	by	IN	B-PP	O
successfully	successfully	RB	B-VP	O
enriching	enrich	VBG	I-VP	O
the	the	DT	B-NP	O
siRNA	siRNA	NN	I-NP	B-DNA
sequences	sequence	NNS	I-NP	I-DNA
into	into	IN	B-PP	O
trans-	trans-	NN	B-NP	O

formed	form	VBN	B-NP	O
matrices	matrix	NNS	I-NP	O
incorporating	incorporate	VBG	B-VP	O
the	the	DT	B-NP	O
effective	effective	JJ	I-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
and	and	CC	O	O
predicting	predict	VBG	B-VP	O
the	the	DT	O	O


A	A	DT	B-NP	O
Novel	Novel	NNP	I-NP	O
Framework	Framework	NNP	I-NP	O
to	to	TO	B-VP	O
Improve	Improve	VB	I-VP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NN	I-NP	O
Prediction	Prediction	NN	I-NP	O


411	411	CD	B-NP	O


siRNA	siRNA	NN	B-NP	O
knockdown	knockdown	JJ	I-NP	O
efficacy	efficacy	NN	I-NP	O
by	by	IN	B-PP	O
bilinear	bilinear	JJ	B-NP	O
tensor	tensor	NN	I-NP	O
regression.	regression.	NN	I-NP	O
The	The	DT	B-NP	O
experimental	experimental	JJ	I-NP	O
com-	com-	NN	I-NP	O

parative	parative	JJ	B-NP	O
evaluation	evaluation	NN	I-NP	O
on	on	IN	B-PP	O
commonly	commonly	RB	B-NP	O
used	use	VBN	I-NP	O
datasets	dataset	NNS	I-NP	O
with	with	IN	B-PP	O
standard	standard	JJ	B-NP	O
evaluation	evaluation	NN	I-NP	O
proce-	proce-	NN	I-NP	O

dure	dure	NN	B-NP	O
in	in	IN	B-PP	O
different	different	JJ	B-NP	O
contexts	context	NNS	I-NP	O
shows	show	VBZ	B-VP	O
that	that	IN	B-SBAR	O
the	the	DT	B-NP	O
proposed	propose	VBN	I-NP	O
framework	framework	NN	I-NP	O
and	and	CC	O	O
corresponding	correspond	VBG	B-VP	O

methods	method	NNS	B-NP	O
achieved	achieve	VBD	B-VP	O
better	good	JJR	B-NP	O
results	result	NNS	I-NP	O
than	than	IN	B-PP	O
most	most	JJS	B-NP	O
existing	exist	VBG	I-NP	O
methods	method	NNS	I-NP	O
for	for	IN	B-PP	O
doing	do	VBG	B-VP	O
the	the	DT	B-NP	O
same	same	JJ	I-NP	O

task.	task.	LS	B-NP	O
One	One	CD	I-NP	O
significant	significant	JJ	I-NP	O
feature	feature	NN	I-NP	O
of	of	IN	B-PP	O
the	the	DT	B-NP	O
proposed	propose	VBN	I-NP	O
framework	framework	NN	I-NP	O
is	be	VBZ	B-VP	O
it	it	PRP	B-NP	O
can	can	MD	B-VP	O
be	be	VB	I-VP	O
easily	easily	RB	I-VP	O
extended	extend	VBN	I-VP	O

when	when	WRB	B-ADVP	O
new	new	JJ	B-NP	O
design	design	NN	I-NP	O
rules	rule	NNS	I-NP	O
are	be	VBP	B-VP	O
discovered	discover	VBN	I-VP	O
as	as	RB	B-CONJP	O
well	well	RB	I-CONJP	O
as	as	IN	I-CONJP	O
more	more	JJR	B-NP	O
siRNAs	siRNA	NNS	I-NP	O
are	be	VBP	B-VP	O
analyzed	analyze	VBN	I-VP	O
by	by	IN	B-PP	O
em-	em-	NN	B-NP	O

pirical	pirical	JJ	B-NP	O
works	work	NNS	I-NP	O
.	.	.	O	O


References	Reference	NNS	B-NP	O


1.	1.	LS	B-LST	O
Amarzguioui	Amarzguioui	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Prydz	Prydz	NNP	B-NP	O
,	,	,	O	O
H.	H.	NNP	B-NP	O
:	:	:	O	O
An	An	DT	B-NP	O
algorithm	algorithm	NN	I-NP	O
for	for	IN	B-PP	O
selection	selection	NN	B-NP	O
of	of	IN	B-PP	O
functional	functional	JJ	B-NP	O
siRNA	siRNA	NN	I-NP	B-protein
se-	se-	NN	I-NP	I-protein


quences.	quences.	LS	B-LST	O
Biochem.	Biochem.	NNP	B-NP	O
Biophys.	Biophys.	NNP	I-NP	O
Res.	Res.	NNP	I-NP	O
Commun.	Commun.	NNP	I-NP	O
316	316	CD	I-NP	O
(	(	(	O	O
4	4	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
1050–1058	1050–1058	CD	B-NP	O
(	(	(	O	O
2004	2004	CD	B-NP	O
)	)	)	O	O


2.	2.	LS	B-LST	O
Elbashir	Elbashir	NNP	B-NP	O
,	,	,	O	O
S.M.	S.M.	NNP	B-NP	O
,	,	,	O	O
Lendeckel	Lendeckel	NNP	B-NP	O
,	,	,	O	O
W.	W.	NNP	B-NP	O
,	,	,	O	O
Tuschl	Tuschl	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
:	:	:	O	O
RNA	RNA	NN	B-NP	O
interference	interference	NN	I-NP	O
is	be	VBZ	B-VP	O
mediated	mediate	VBN	I-VP	O
by	by	IN	B-PP	O


21–	21–	NN	B-NP	O
and	and	CC	I-NP	O
22–nucleotide	22–nucleotide	NN	I-NP	O
RNAs.	RNAs.	NNP	I-NP	O
Genes	Genes	NNP	I-NP	O
Dev.	Dev.	NNP	I-NP	O
15	15	CD	I-NP	O
,	,	,	I-NP	O
188–200	188–200	CD	I-NP	O
(	(	(	O	O
2001	2001	CD	B-NP	O
)	)	)	O	O


3.	3.	LS	B-LST	O
Harborth	Harborth	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Elbashir	Elbashir	NNP	B-NP	O
,	,	,	O	O
S.M.	S.M.	NNP	B-NP	O
,	,	,	O	O
Vandenburgh	Vandenburgh	NNP	B-NP	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
,	,	,	O	O
Manninga	Manninga	NNP	B-NP	O
,	,	,	O	O
H.	H.	NNP	B-NP	O
,	,	,	O	O
Scaringe	Scaringe	NNP	B-NP	O
,	,	,	O	O
S.A.	S.A.	NNP	B-NP	O
,	,	,	O	O
We-	We-	NNP	B-NP	O

ber	ber	RB	B-ADVP	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
,	,	,	O	O
Tuschl	Tuschl	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
:	:	:	O	O
Sequence	Sequence	NN	B-NP	O
,	,	,	O	O
chemical	chemical	NN	B-NP	O
,	,	,	O	O
and	and	CC	O	O
structural	structural	JJ	B-NP	O
variation	variation	NN	I-NP	O
of	of	IN	B-PP	O
small	small	JJ	B-NP	O
interfering	interfere	VBG	I-NP	O

RNAs	RNA	NNS	B-NP	B-RNA
and	and	CC	O	O
short	short	JJ	B-NP	O
hairpin	hairpin	NN	I-NP	O
RNAs	RNA	NNS	I-NP	O
and	and	CC	O	O
the	the	DT	B-NP	O
effect	effect	NN	I-NP	O
on	on	IN	B-PP	O
mammalian	mammalian	JJ	B-NP	B-DNA
gene	gene	NN	I-NP	I-DNA
silencing.	silencing.	NN	I-NP	I-DNA
Anti-	Anti-	.	O	I-DNA

sense	sense	NN	B-NP	O
Nucleic	Nucleic	NNP	I-NP	O
Acid	Acid	NNP	I-NP	O
Drug	Drug	NNP	I-NP	O
Dev.	Dev.	NNP	I-NP	O
13	13	CD	I-NP	O
,	,	,	I-NP	O
83–105	83–105	CD	I-NP	O
(	(	(	O	O
2003	2003	CD	B-NP	O
)	)	)	O	O


4.	4.	LS	B-LST	O
Hsieh	Hsieh	NNP	B-NP	O
,	,	,	O	O
A.C.	A.C.	NNP	B-NP	O
,	,	,	O	O
Bo	Bo	NNP	B-NP	O
,	,	,	O	O
R.	R.	NNP	B-NP	O
,	,	,	O	O
Manola	Manola	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Vazquez	Vazquez	NNP	B-NP	O
,	,	,	O	O
F.	F.	NNP	B-NP	O
,	,	,	O	O
Bare	Bare	NNP	B-NP	O
,	,	,	O	O
O.	O.	NNP	B-NP	O
,	,	,	O	O
Khvorova	Khvorova	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Scaringe	Scaringe	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O

Sellers	Seller	NNS	B-NP	O
,	,	,	O	O
W.R.	W.R.	NNP	B-NP	O
:	:	:	O	O
A	A	DT	B-NP	O
library	library	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
duplexes	duplex	NNS	I-NP	O
targeting	target	VBG	B-VP	O
the	the	DT	B-NP	O
phosphoinositide	phosphoinositide	NN	I-NP	B-protein
3-kinase	3-kinase	NN	I-NP	I-protein

pathway	pathway	NN	B-NP	O
:	:	:	O	O
Determinants	Determinant	NNS	B-NP	O
of	of	IN	B-PP	O
gene	gene	NN	B-NP	O
silencing	silencing	NN	I-NP	O
for	for	IN	B-PP	O
use	use	NN	B-NP	O
in	in	IN	B-PP	O
cell-based	cell-based	JJ	B-NP	O
screens.	screens.	NN	I-NP	O
Nucleic	Nucleic	JJ	I-NP	O
Acids	Acid	NNS	I-NP	O

Res.	Res.	NN	B-NP	O
32	32	CD	I-NP	O
(	(	(	O	O
3	3	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
893–901	893–901	CD	B-NP	O
(	(	(	O	O
2004	2004	CD	B-NP	O
)	)	)	O	O


5.	5.	LS	B-LST	O
Huesken	Huesken	NNP	B-NP	O
,	,	,	O	O
D.	D.	NNP	B-NP	O
,	,	,	O	O
Lange	Lange	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Mickanin	Mickanin	NNP	B-NP	O
,	,	,	O	O
C.	C.	NNP	B-NP	O
,	,	,	O	O
Weiler	Weiler	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Asselbergs	Asselbergs	NNP	B-NP	O
,	,	,	O	O
F.	F.	NNP	B-NP	O
,	,	,	O	O
Warner	Warner	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Mellon	Mellon	NNP	B-NP	O
,	,	,	O	O

B.	B.	NNP	B-NP	O
,	,	,	O	O
Engel	Engel	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
Rosenberg	Rosenberg	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Cohen	Cohen	NNP	B-NP	O
,	,	,	O	O
D.	D.	NNP	B-NP	O
,	,	,	O	O
Labow	Labow	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Reinhardt	Reinhardt	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Natt	Natt	NNP	B-NP	O
,	,	,	O	O
F.	F.	NNP	B-NP	O
,	,	,	O	O
Hall	Hall	NNP	B-NP	O
,	,	,	O	O

J.	J.	NNP	B-NP	O
:	:	:	O	O
Design	Design	NNP	B-NP	O
of	of	IN	B-PP	O
a	a	DT	B-NP	O
Genome-Wide	Genome-Wide	JJ	I-NP	O
siRNA	siRNA	NN	I-NP	O
Library	Library	NN	I-NP	O
Using	Use	VBG	B-VP	O
an	an	DT	B-NP	O
Artificial	Artificial	NNP	I-NP	O
Neural	Neural	NNP	I-NP	O
Network	Network	NNP	I-NP	O
.	.	.	O	O

Nature	Nature	NNP	B-NP	O
Biotechnology	Biotechnology	NNP	I-NP	O
23	23	CD	I-NP	O
(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
955–1001	955–1001	CD	B-NP	O
(	(	(	O	O
2005	2005	CD	B-NP	O
)	)	)	O	O


6.	6.	LS	B-LST	O
Ichihara	Ichihara	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Murakumo	Murakumo	NNP	B-NP	O
,	,	,	O	O
Y.	Y.	NNP	B-NP	O
,	,	,	O	O
Masuda	Masuda	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Matsuura	Matsuura	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
,	,	,	O	O
Asai	Asai	NNP	B-NP	O
,	,	,	O	O
N.	N.	NNP	B-NP	O
,	,	,	O	O
Jijiwa	Jijiwa	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O

Ishida	Ishida	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Shinmi	Shinmi	NNP	B-NP	O
,	,	,	O	O
J.	J.	NNP	B-NP	O
,	,	,	O	O
Yatsuya	Yatsuya	NNP	B-NP	O
,	,	,	O	O
H.	H.	NNP	B-NP	O
,	,	,	O	O
Qiao	Qiao	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
et	et	FW	B-ADVP	O
al.	al.	FW	I-ADVP	O
:	:	:	O	O
Thermodynamic	Thermodynamic	JJ	B-NP	O
instability	instability	NN	I-NP	O

of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
duplex	duplex	NN	I-NP	O
is	be	VBZ	B-VP	O
a	a	DT	B-NP	O
prerequisite	prerequisite	NN	I-NP	O
for	for	IN	B-PP	O
dependable	dependable	JJ	B-NP	O
prediction	prediction	NN	I-NP	O
of	of	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
activities	activity	NNS	I-NP	O
.	.	.	O	O

Nucleic	Nucleic	JJ	B-NP	O
Acids	Acids	NNP	I-NP	O
Res.	Res.	NNP	I-NP	O
35	35	CD	I-NP	O
(	(	(	O	O
8	8	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
e123	e123	NN	B-NP	B-protein
(	(	(	O	O
2007	2007	CD	B-NP	O
)	)	)	O	O


7.	7.	CD	B-NP	O
Jagla	Jagla	NNP	I-NP	O
,	,	,	O	O
B.	B.	NNP	B-NP	O
,	,	,	O	O
Aulner	Aulner	NNP	B-NP	O
,	,	,	O	O
N.	N.	NNP	B-NP	O
,	,	,	O	O
Kelly	Kelly	NNP	B-NP	O
,	,	,	O	O
P.D.	P.D.	NNP	B-NP	O
,	,	,	O	O
Song	Song	NNP	B-NP	O
,	,	,	O	O
D.	D.	NNP	B-NP	O
,	,	,	O	O
Volchuk	Volchuk	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Zatorski	Zatorski	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Shum	Shum	NNP	B-NP	O
,	,	,	O	O

D.	D.	NNP	B-NP	O
,	,	,	O	O
Mayer	Mayer	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
,	,	,	O	O
De	De	NNP	B-NP	O
Angelis	Angelis	NNP	I-NP	O
,	,	,	O	O
D.A.	D.A.	NNP	B-NP	O
,	,	,	O	O
Ouerfelli	Ouerfelli	NNP	B-NP	O
,	,	,	O	O
O.	O.	NNP	B-NP	O
,	,	,	O	O
Rutishauser	Rutishauser	NNP	B-NP	O
,	,	,	O	O
U.	U.	NNP	B-NP	O
,	,	,	O	O
Rothman	Rothman	NNP	B-NP	O
,	,	,	O	O
J.E.	J.E.	NNP	B-NP	O
:	:	:	O	O

Sequence	Sequence	NN	B-NP	O
characteristics	characteristic	NNS	I-NP	O
of	of	IN	B-PP	O
functional	functional	JJ	B-NP	O
siRNAs.	siRNAs.	NN	I-NP	O
RNA	RNA	NN	I-NP	O
11	11	CD	I-NP	O
(	(	(	O	O
6	6	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
864–872	864–872	CD	B-NP	O
(	(	(	O	O
2005	2005	CD	B-NP	O
)	)	)	O	O


8.	8.	CD	B-NP	O
Qiu	Qiu	NNP	I-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
Lane	Lane	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
:	:	:	O	O
A	A	NNP	B-NP	O
Framework	Framework	NNP	I-NP	O
for	for	IN	B-PP	O
Multiple	Multiple	NNP	B-NP	O
Kernel	Kernel	NNP	I-NP	O
Support	Support	NNP	I-NP	O
Vector	Vector	NNP	I-NP	O
Regression	Regression	NNP	I-NP	O

and	and	CC	O	O
Its	Its	PRP$	B-NP	O
Applications	Application	NNS	I-NP	O
to	to	TO	B-PP	O
siRNA	siRNA	NN	B-NP	O
Efficacy	Efficacy	NNP	I-NP	O
Prediction.	Prediction.	NNP	I-NP	O
IEEE/ACM	IEEE/ACM	NNP	I-NP	O
Trans.	Trans.	NNP	I-NP	O
Comput	Comput	NNP	I-NP	O
.	.	.	O	O

Biology	Biology	NNP	B-NP	O
Bioinform.	Bioinform.	NNP	I-NP	O
6	6	CD	I-NP	O
(	(	(	O	O
2	2	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
190–199	190–199	CD	B-NP	O
(	(	(	O	O
2009	2009	CD	B-NP	O
)	)	)	O	O


9.	9.	CD	B-NP	O
Ren	Ren	NNP	I-NP	O
,	,	,	O	O
Y.	Y.	NNP	B-NP	O
,	,	,	O	O
Gong	Gong	NNP	B-NP	O
,	,	,	O	O
W.	W.	NNP	B-NP	O
,	,	,	O	O
Xu	Xu	NNP	B-NP	O
,	,	,	O	O
Q.	Q.	NNP	B-NP	O
,	,	,	O	O
Zheng	Zheng	NNP	B-NP	O
,	,	,	O	O
X.	X.	NNP	B-NP	O
,	,	,	O	O
Lin	Lin	NNP	B-NP	B-protein
,	,	,	O	O
D.	D.	NNP	B-NP	O
,	,	,	O	O
et	et	FW	B-ADVP	O
al.	al.	FW	I-ADVP	O
:	:	:	O	O
siRecords	siRecord	NNS	B-NP	O
:	:	:	O	O
An	An	DT	B-NP	O
extensive	extensive	JJ	I-NP	O

database	database	NN	B-NP	O
of	of	IN	B-PP	O
mammalian	mammalian	JJ	B-NP	B-RNA
siRNAs	siRNA	NNS	I-NP	I-RNA
with	with	IN	B-PP	O
efficacy	efficacy	NN	B-NP	O
ratings.	ratings.	NN	I-NP	O
Bioinformatics	Bioinformatic	NNS	I-NP	O
22	22	CD	B-NP	O
,	,	,	I-NP	O
1027–1028	1027–1028	CD	I-NP	O

(	(	(	O	O
2006	2006	CD	B-NP	O
)	)	)	O	O


10.	10.	LS	B-LST	O
Reynolds	Reynolds	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O
Leake	Leake	NNP	B-NP	O
,	,	,	O	O
D.	D.	NNP	B-NP	O
,	,	,	O	O
Boese	Boese	NNP	B-NP	O
,	,	,	O	O
Q.	Q.	NNP	B-NP	O
,	,	,	O	O
Scaringe	Scaringe	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
Marshall	Marshall	NNP	B-NP	O
,	,	,	O	O
W.S.	W.S.	NNP	B-NP	O
,	,	,	O	O
Khvorova	Khvorova	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
:	:	:	O	O

Rational	Rational	JJ	B-NP	O
siRNA	siRNA	NN	I-NP	O
design	design	NN	I-NP	O
for	for	IN	B-PP	O
RNA	RNA	NN	B-NP	O
interference.	interference.	NN	I-NP	O
Nat.	Nat.	NNP	I-NP	O
Biotechnol.	Biotechnol.	NNP	I-NP	O
22	22	CD	I-NP	O
(	(	(	O	O
3	3	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
326–330	326–330	CD	B-NP	O

(	(	(	O	O
2004	2004	CD	B-NP	O
)	)	)	O	O


11.	11.	LS	B-LST	O
Scherer	Scherer	NNP	B-NP	O
,	,	,	O	O
L.J.	L.J.	NNP	B-NP	O
,	,	,	O	O
Rossi	Rossi	NNP	B-NP	O
,	,	,	O	O
J.J.	J.J.	NNP	B-NP	O
:	:	:	O	O
Approaches	Approaches	NNP	B-NP	O
for	for	IN	B-PP	O
the	the	DT	B-NP	O
sequence-specific	sequence-specific	NN	I-NP	O
knockdown	knockdown	NN	I-NP	O
of	of	IN	B-PP	O


mRNA.	mRNA.	NNP	B-NP	O
Nat.	Nat.	NNP	I-NP	O
Biotechnol.	Biotechnol.	NNP	I-NP	O
21	21	CD	I-NP	O
,	,	,	I-NP	O
1457–1465	1457–1465	CD	I-NP	O
(	(	(	O	O
2003	2003	CD	B-NP	O
)	)	)	O	O


12.	12.	CD	B-NP	O
Sciabola	Sciabola	NNP	I-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
Cao	Cao	NNP	B-NP	O
,	,	,	O	O
Q.	Q.	NNP	B-NP	O
,	,	,	O	O
Orozco	Orozco	NNP	B-NP	O
,	,	,	O	O
M.	M.	NNP	B-NP	O
,	,	,	O	O
Faustino	Faustino	NNP	B-NP	O
,	,	,	O	O
I.	I.	NNP	B-NP	O
,	,	,	O	O
Stanton	Stanton	NNP	B-NP	O
,	,	,	O	O
R.V.	R.V.	NNP	B-NP	O
:	:	:	O	O
Improved	Improve	VBD	B-VP	O
nucleic	nucleic	JJ	B-ADJP	O

acid	acid	NN	B-NP	O
descriptors	descriptor	NNS	I-NP	O
for	for	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
efficacy	efficacy	NN	I-NP	O
prediction.	prediction.	VBD	B-VP	O
Nucl.	Nucl.	NNP	B-NP	O
Acids	Acids	NNP	I-NP	O
Res.	Res.	NNP	I-NP	O
41	41	CD	I-NP	O
(	(	(	O	O
3	3	CD	B-NP	O
)	)	)	O	O
,	,	,	O	O
1383–1394	1383–1394	CD	B-NP	O

(	(	(	O	O
2013	2013	CD	B-NP	O
)	)	)	O	O


412	412	CD	B-NP	O


B.T.	B.T.	NNP	B-NP	O
Ngoc	Ngoc	NNP	I-NP	O


13.	13.	LS	B-LST	O
Shabalina	Shabalina	NNP	B-NP	O
,	,	,	O	O
S.A.	S.A.	NNP	B-NP	O
,	,	,	O	O
Spiridonov	Spiridonov	NNP	B-NP	O
,	,	,	O	O
A.N.	A.N.	NNP	B-NP	O
,	,	,	O	O
Ogurtsov	Ogurtsov	NNP	B-NP	O
,	,	,	O	O
A.Y.	A.Y.	NNP	B-NP	O
:	:	:	O	O
Computational	Computational	JJ	B-NP	O
models	model	NNS	I-NP	O
with	with	IN	B-PP	O

thermodynamic	thermodynamic	JJ	B-NP	O
and	and	CC	I-NP	O
composition	composition	NN	I-NP	O
features	feature	NNS	I-NP	O
improve	improve	VBP	B-VP	O
siRNA	siRNA	NN	B-NP	O
design.	design.	NN	I-NP	O
BMC	BMC	NNP	I-NP	O
Bioinfor-	Bioinfor-	NNP	I-NP	O

matics	matic	NNS	B-NP	O
7	7	CD	I-NP	O
,	,	,	O	O
65	65	CD	B-NP	O
(	(	(	O	O
2006	2006	CD	B-NP	O
)	)	)	O	O


14.	14.	LS	B-LST	O
Takasaki	Takasaki	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
:	:	:	O	O
Methods	Method	NNS	B-NP	O
for	for	IN	B-PP	O
Selecting	Select	VBG	B-VP	O
Effective	Effective	JJ	B-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
Target	Target	NN	I-NP	I-DNA
Sequences	Sequence	NNS	I-NP	I-DNA
Using	Use	VBG	B-VP	O
a	a	DT	B-NP	O

Variety	Variety	NNP	B-NP	O
of	of	IN	B-PP	O
Statistical	Statistical	NNP	B-NP	O
and	and	CC	O	O
Analytical	Analytical	NNP	B-NP	O
Techniques.	Techniques.	NNP	I-NP	O
Methods	Methods	NNP	I-NP	O
Mol.	Mol.	NNP	I-NP	O
Biol.	Biol.	NNP	I-NP	O
942	942	CD	I-NP	O
,	,	,	I-NP	O
17–55	17–55	CD	I-NP	O

(	(	(	O	O
2013	2013	CD	B-NP	O
)	)	)	O	O


15.	15.	LS	B-LST	O
Ui-Tei	Ui-Tei	NNP	B-NP	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
,	,	,	O	O
Naito	Naito	NNP	B-NP	O
,	,	,	O	O
Y.	Y.	NNP	B-NP	O
,	,	,	O	O
Takahashi	Takahashi	NNP	B-NP	O
,	,	,	O	O
F.	F.	NNP	B-NP	O
,	,	,	O	O
Haraguchi	Haraguchi	NNP	B-NP	O
,	,	,	O	O
T.	T.	NNP	B-NP	O
,	,	,	O	O
Ohki-Hamazaki	Ohki-Hamazaki	NNP	B-NP	O
,	,	,	O	O
H.	H.	NNP	B-NP	O
,	,	,	O	O
Juni	Juni	NNP	B-NP	O
,	,	,	O	O
A.	A.	NNP	B-NP	O
,	,	,	O	O

Ueda	Ueda	NNP	B-NP	O
,	,	,	O	O
R.	R.	NNP	B-NP	O
,	,	,	O	O
Saigo	Saigo	NNP	B-NP	O
,	,	,	O	O
K.	K.	NNP	B-NP	O
:	:	:	O	O
Guidelines	Guideline	NNS	B-NP	O
for	for	IN	B-PP	O
the	the	DT	B-NP	O
selection	selection	NN	I-NP	O
of	of	IN	B-PP	O
highly	highly	RB	B-NP	O
effective	effective	JJ	I-NP	B-DNA
siRNA	siRNA	NN	I-NP	I-DNA
sequences	sequence	NNS	I-NP	I-DNA

for	for	IN	B-PP	O
mammalian	mammalian	JJ	B-NP	O
and	and	CC	I-NP	O
chick	chick	NN	I-NP	O
RNA	RNA	NN	I-NP	O
interference.	interference.	NN	I-NP	O
Nucleic	Nucleic	NNP	I-NP	O
Acids	Acids	NNP	I-NP	O
Res.	Res.	NNP	I-NP	O
32	32	CD	I-NP	O
,	,	,	I-NP	O
936–948	936–948	CD	I-NP	O
(	(	(	O	O
2004	2004	CD	B-NP	O
)	)	)	O	O

16.	16.	LS	B-LST	O
Vert	Vert	NNP	B-NP	O
,	,	,	O	O
J.P.	J.P.	NNP	B-NP	O
,	,	,	O	O
Foveau	Foveau	NNP	B-NP	O
,	,	,	O	O
N.	N.	NNP	B-NP	O
,	,	,	O	O
Lajaunie	Lajaunie	NNP	B-NP	O
,	,	,	O	O
C.	C.	NNP	B-NP	O
,	,	,	O	O
Vandenbrouck	Vandenbrouck	NNP	B-NP	O
,	,	,	O	O
Y.	Y.	NNP	B-NP	O
:	:	:	O	O
An	An	DT	B-NP	O
accurate	accurate	JJ	I-NP	O
and	and	CC	I-NP	O
inter-	inter-	JJ	I-NP	O


pretable	pretable	JJ	B-NP	O
model	model	NN	I-NP	O
for	for	IN	B-PP	O
siRNA	siRNA	NN	B-NP	O
efficacy	efficacy	NN	I-NP	O
prediction.	prediction.	VBD	B-VP	O
BMC	BMC	NNP	B-NP	B-cell_type
Bioinformatics	Bioinformatics	NNP	I-NP	O
7	7	CD	I-NP	O
,	,	,	I-NP	O
520	520	CD	I-NP	O
(	(	(	O	O
2006	2006	CD	B-NP	O
)	)	)	O	O


17.	17.	CD	B-NP	O
Vickers	Vickers	NNP	I-NP	O
,	,	,	O	O
T.A.	T.A.	NNP	B-NP	O
,	,	,	O	O
Koo	Koo	NNP	B-NP	O
,	,	,	O	O
S.	S.	NNP	B-NP	O
,	,	,	O	O
Bennett	Bennett	NNP	B-NP	O
,	,	,	O	O
C.F.	C.F.	NNP	B-NP	O
,	,	,	O	O
Crooke	Crooke	NNP	B-NP	O
,	,	,	O	O
S.T.	S.T.	NNP	B-NP	O
,	,	,	O	O
Dean	Dean	NNP	B-NP	O
,	,	,	O	O
N.M.	N.M.	NNP	B-NP	O
,	,	,	O	O
Baker	Baker	NNP	B-NP	O
,	,	,	O	O
B.F.	B.F.	NNP	B-NP	O
:	:	:	O	O
Effi-	Effi-	NN	B-NP	O

cient	cient	JJ	B-NP	O
reduction	reduction	NN	I-NP	O
of	of	IN	B-PP	O
target	target	NN	B-NP	B-RNA
RNAs	RNA	NNS	I-NP	I-RNA
by	by	IN	B-PP	O
small	small	JJ	B-NP	O
interfering	interfere	VBG	I-NP	O
RNA	RNA	NN	I-NP	O
and	and	CC	I-NP	O
RNase	RNase	NN	I-NP	B-protein
H-dependent	H-dependent	JJ	B-ADJP	O

antisense	antisense	JJ	B-NP	O
agents.	agents.	NN	I-NP	O
A	A	NN	I-NP	O
Comparative	Comparative	NNP	I-NP	O
Analysis.	Analysis.	NNP	I-NP	O
J.	J.	NNP	I-NP	O
Biol.	Biol.	NNP	I-NP	O
Chem.	Chem.	NNP	I-NP	O
278	278	CD	I-NP	O
,	,	,	I-NP	O
7108–7118	7108–7118	CD	I-NP	O
(	(	(	O	O
2003	2003	CD	B-NP	O
)	)	)	O	O