Unexpected behaviour with stretching w/o anchor

[mschubert]

Hi,

I'm trying to stretch a genomic region using different anchoring. I also tried without anchoring, which yielded the (unexpected) result below:

> x = makeGRangesFromDataFrame(data.frame(
    seqnames=1:2, start=c(10,100), end=c(200,300), strand=c('+','-')))

> x
      seqnames     ranges strand
         <Rle>  <IRanges>  <Rle>
  [1]        1    10-200      +
  [2]        2   100-300      -

> stretch(x, 2)
      seqnames     ranges strand
         <Rle>  <IRanges>  <Rle>
  [1]        1   104-106      +
  [2]        2   199-201      -

Why does it change the regions this way? (anchored stretching works as expected)

Using IRanges_2.14.10 and plyranges_1.0.3 (current Bioconductor release).

[sa-lee]

It looks like the default is to stretch out from the centre, , which is not what my documentation says, so this is at least unexpected in that regard.

What output would expect here? Stretching by 2 away from the start and end?

> x %>% mutate(start = start - 2L, end = end + 2L)
GRanges object with 2 ranges and 0 metadata columns:
      seqnames    ranges strand
         <Rle> <IRanges>  <Rle>
  [1]        1     8-202      +
  [2]        2    98-302      -
  -------
  seqinfo: 2 sequences from an unspecified genome; no seqlengths

[lawremi]

Probably should behave however the start<-(), end()<- etc setters behave now.

[sa-lee]

mutate uses those setters behind the scenes so the behaviour should be consistent.

[sa-lee]

wondering if we could refactor here to have stretch just work inside mutate like we have been discussing in #47

[mschubert]

@sa-lee I would expect stretch(., 0) to be a no-op, positive numbers to increase the overall range and negative to decrease overall range.

Personally, I wouldn't mind an error (or warning + default anchor) if the interval is not anchored.

[lawremi]

Agreed, I would have expected stretch_center() (the internal function corresponding to this case) to be like the +,Ranges() in IRanges: it would add half the amount to each end.

[mschubert]

Note that this also happens for:

> stretch(anchor_center(x), 2)
      seqnames     ranges strand
         <Rle>  <IRanges>  <Rle>
  [1]        1   104-106      +
  [2]        2   199-201      -

(start, end, 5p, 3p anchors work as expected)

[sa-lee]

Okey I'll change the default/centering to be like +, Ranges for now. I am also leaning towards your suggestion of having stretch throw a warning for no anchoring.

[mschubert]

I didn't realize at before, but I'm also confused about anchor_end:

> stretch(anchor_end(x), 2)
      seqnames     ranges strand
         <Rle>  <IRanges>  <Rle>
  [1]        1    12-200      +
  [2]        2   102-300      -

This seems to shorten rather than stretch?

[sa-lee]

Yep that looks wrong to me too. It should go the the other way. Turns out there was a wayward + in that code.

[mschubert]

🎉