USI reported by quantms for DIANN experiments.

[ypriverol]

Description of the bug

@WangHong007 @zprobot @daichengxin:

I have been recently testing multiple USIs generated by quantms for DIA experiments. A little background about the problem, USIs are a way to reference directly to the scan and spectrum that was used to identify the spectrum, in someway is the fundamental evidence on DDA identification; here is an example.

All DDA search engines keep track of the scan that was used to identify the spectrum. However, in DIA experiments other features are also relevant and DIANN do not trace in the output files of the scan number that was used to identify the peptide. In quantms, we have implemented a logic to "find" for every peptide the scan number used to identify the peptide. @zprobot provided all the USIs for reanalysis PXD019909; however, when we were doing a visualization of multiple USIs they look like random USI meaning the spectrum looks like do not correspond to the given peptide, see example.

I propose the following follow up:

• @WangHong007 can you explain in a comment on this thread in details how the scan are retrieved. Probably @vdemichev can double-check that our logic is correct.
• @zprobot can you provide for other reanalyses the corresponding USIs and the Posterior error probabilities. The idea is that we can check for the best IDs (lower PEP) in their USIs. It is important to have the USIs and PEP for other projects (reanalyses) to see if the problems are related with the specific project.
• Double check that for this project PXD019909 the USIs are well annotated @WangHong007 @zprobot. If they are well annotated, and we check USIs for good IDs (lower PEP) we can try to discuss with @vdemichev what is the problem on this project.

Please let me know if you need more discussion.

Additional examples that looks wrong:

• https://www.ebi.ac.uk/pride/archive/usi?usi=mzspec:PXD019909:20180914_QE8_nLC0_BDA_SA_DIA_Keratinocytes_NN002:scan:7150:GKQEEEKPGEEK/2&resultType=FULL
• https://www.ebi.ac.uk/pride/archive/usi?usi=mzspec:PXD019909:20180914_QE8_nLC0_BDA_SA_DIA_Keratinocytes_NN002:scan:6918:HSQAGQGQSEGSR/2&resultType=FULL

Command used and terminal output

No response

Relevant files

No response

System information

No response

[vdemichev]

In non-tims data the apex RT reported by DIA-NN should uniquely identify the scan.
Why do those examples look wrong? They have some fragments annotated.

[ypriverol]

Thanks, @vdemichev, for responding. We were having a look to some of them and saw not to many fragments, for example https://www.ebi.ac.uk/pride/archive/usi?usi=mzspec%3APXD019909%3A20180914_QE8_nLC0_BDA_SA_DIA_Keratinocytes_NN002%3Ascan%3A7150%3AGKQEEEKPGEEK%2F2&resultType=FULL. If you use in the viewer mass type mono and 40 ppm you will see only 4 b-ions and 2 y-ions, the Proline for example is not even identified. I'm not used to see DIA spectra in viz tools, my question is:

• Is this kind of number of fragment annotations normally? It could be a good ID only with 6 ions.
• Is this because apart from the fragment ions, other info is used as precursor mz and rt?

USIs and viz of spectra in DDA is quite common, see this https://www.ebi.ac.uk/pride/archive/usi?usi=mzspec:PXD000561:Adult_Frontalcortex_bRP_Elite_85_f09:scan:17555:VLHPLEGAVVIIFK/2 we are exploring and checking if this may sense for DIA to write some guidelines about DIA USIs. Your feedback is more than welcome. BTW, we are the only one's generating USIs for DIA & DIANN experiments.

[jpfeuffer]

Hi all,

I am currently looking into similar visualizations from quantms DIA-NN output.
I think DIA-NN uses much more than just a single MS2 spectrum for identification.
What you ideally want to display is a view as in Skyline or OpenSWATH, where you visualize the co-elution of fragment ions for a peptide along RT (potentially with competing groups from other peptides). This means for every found fragment you would want start and end RT. I am in the process of checking if this information is available and how to extract it.

DDA-like PSM visualization at the apex can be a supplementary view but might be very convoluted.
Looking forward to more input from @vdemichev, though!

[WangHong007]

Here the logic how we match diann report to ms_info.tsv:

quantms/bin/diann_convert.py

Lines 838 to 858 in fd20331

	for n, group in report.groupby(["Run"]):
	if isinstance(n, tuple) and len(n) == 1:
	# This is here only to support versions of pandas where the groupby
	# key is a tuple.
	# related: https://github.com/pandas-dev/pandas/pull/51817
	n = n[0]
	file = __find_info(folder, n)
	target = pd.read_csv(file, sep="\t")
	group.sort_values(by="RT.Start", inplace=True)
	target = target[["Retention_Time", "SpectrumID", "Exp_Mass_To_Charge"]]
	target.columns = ["RT.Start", "opt_global_spectrum_reference", "exp_mass_to_charge"]
	# Standardize spectrum identifier format for bruker data
	if type(target.loc[0, "opt_global_spectrum_reference"]) != str:
	target.loc[:, "opt_global_spectrum_reference"] = "scan=" + target.loc[
	:, "opt_global_spectrum_reference"
	].astype(str)
	# TODO seconds returned from precursor.getRT()
	target.loc[:, "RT.Start"] = target.apply(lambda x: x["RT.Start"] / 60, axis=1)
	out_mztab_PSH = pd.concat([out_mztab_PSH, pd.merge_asof(group, target, on="RT.Start", direction="nearest")])

Basically, we choose 'RT.Start' (seconds as unit) as reteintion time in the report and changed seconds to minutes, at last we merge report and ms_info on column 'RT.start' to make sure matching nearest reteintion times.

[vdemichev]

I would suggest to use RT instead of RT.Start

[ypriverol]

Hi all,

I am currently looking into similar visualizations from quantms DIA-NN output. I think DIA-NN uses much more than just a single MS2 spectrum for identification. What you ideally want to display is a view as in Skyline or OpenSWATH, where you visualize the co-elution of fragment ions for a peptide along RT (potentially with competing groups from other peptides). This means for every found fragment you would want start and end RT. I am in the process of checking if this information is available and how to extract it.

DDA-like PSM visualization at the apex can be a supplementary view but might be very convoluted. Looking forward to more input from @vdemichev, though!

@jpfeuffer can you provide some ideas and visualization about examples it. I don't have access to Skyline?

[jpfeuffer]

Me neither but this is from their tutorial:

The top right is I think similar to what you are currently trying to do (probably at the RT apex [= RT] not RT.start or RT.end).
You can also compare with the actual expected library intensities in a mirror plot or something.

The left charts are XICs from precursor isotopes and fragment ions. Those will require a range of MS2 spectra to extract from.

I guess for the use-case of a USI, your approach may be enough if you switch to RT instead of RT.start.

[WangHong007]

Hi @jpfeuffer @vdemichev, in the discussion above, you mentioned using apex RT to match the RT column in the diann report, and I have an additional question: When we use pyopenms to process mass spectrometry files (such as mzMLs), the getRT() function takes the scan start time in the mzml file as the retention time and converts it to seconds. The retention time here is start RT not apex RT, right? How do we get apex RT of peaks?

[jpfeuffer]

I think there might be a confusion between "scan start time" of a spectrum and "elution start time" of a peptide or peptide fragment.

I think you have to compare "RT" (= apex of peptide elution) from DIA-NN with "getRT()" (="scan [start] time") from pyopenms.
Correct me if I am wrong.