Processing DIA-NN output prior to FDP estimation

[a-indra]

Hi FDRBench developers,
This is a rather rudimentary question relative to the others on this forum as I am new to the tool.

I have used FDRBench to generate both* a shuffled (Human) and foreign (Human + A. thaliana) entrapment database, at the protein level. This was then used in Fragpipe using one of the in-built DIA workflows (DIA_Speclib_Quant). This workflow generates reports pre- and post-DIANN quant, but I am unsure which report is most appropriate for use in FDRBench.

*Edit to clarify: two FASTAs were generated for two independent searches

In order of increasing depth, some candidate outputs containing scoring information include:

• "protein.tsv": per protein reporting: Protein Probability, Top Peptide Probability, Protein Qvalue
• "peptide.tsv": per-peptide matrix reporting: Probability, Qvalue
• "ion.tsv": per-modified peptide matrix reporting: Probability, Qvalue, Expectation
• "psm.tsv" is a per-spectrum matrix reporting: SpectralSim, RTScore, Expectation, Hyperscore, Nextscore, Probability, Qvalue
• "report.tsv" (DIANN-generated) is a comprehensive per-run large matrix that reports many different values including: Q.Value, PEP, Global.Q.Value Lib.Q.Value, PG.Q.Value, PG.PEP, GG.Q.Value, Protein.Q.Value, Global.PG.Q.Value, Lib.PG.Q.Value

The last in the list is the only one generated by DIANN. As I have enabled MBR for DIANN within Fragpipe, I wonder whether that is the most appropriate. However, it is a rather large file (4.65GB). There is also report.parquet at 780 MB.

Can you please advise which output is appropriate, which column should be used as the basis for 'score', and what processing steps you'd recommend prior to use in FDRBench for FDP estimation?

Thank you for this tool and thank you in advance for any advice.
Anuk

[wenbostar]

Hi @fcyu , could you advise which output should be used from the FragPipe's DIA_Speclib_Quant workflow?

[fcyu]

If you want to evaluate the identification FDP, use protein.tsv for protein level and peptide.tsv for peptide level. If you want to evaluate the FDP of the whole workflow (identification+quantification), use report.tsv.

Just in case, after generating the FASTA file with entrapments (without decoys), you need to generate decoys using FragPipe.

Best,

Fengchao

[a-indra]

Thank you both for the prompt response, much appreciated.

In both cases, which column is the appropriate one to use as score for FDRBench?

[fcyu]

It is a little complicated, and, AFAIK, there are no perfect solutions to your question.

1. protein.tsv is filtered with multidimensional FDRs (PSM and protein levels). All the entries in the file have already been filtered at the combination of 1% PSM- and protein-level FDR (by default). So you will only obtain one FDP value rather than an array (with respect to different FDR thresholds). Check how we did it in https://www.nature.com/articles/s41467-025-58728-z, and here (https://github.com/Nesvilab/EntrapBench) are the scripts to do it. It uses FDRBench's equation.
2. report.tsv has already been filtered with a 1% run-wise precursor q-value by default, so you won't get the entries with precursor FDR > 1%. Also, there are multiple FDRs at different levels and run-wise vs global in the report.tsv file. @wenbostar , could you please comment on how the FDRBench paper does with the report.tsv?

My recommendation is that you do not try to get a 2-D plot with FDP vs different FDR thresholds because the thresholds are not a single FDR value and some of the files have already been filtered at pre-defined thresholds. Instead, get a single FDP value given the software's default settings.

Thanks,

Fengchao

[wenbostar]

report.tsv has already been filtered with a 1% run-wise precursor q-value by default, so you won't get the entries with precursor FDR > 1%. Also, there are multiple FDRs at different levels and run-wise vs global in the report.tsv file. @wenbostar , could you please comment on how the FDRBench paper does with the report.tsv?

If just want to estimate FDP for 1% FDR threshold, it's default FDR setting (1%) should work.