Lower IDs with DIA-NN 2.6.1 compared to 2.3.2

[Alpine2048]

232report.log.txt
232report_runs.pdf
261report.log.txt
261report_runs.pdf

Hi Vadim,

I am trying to analyze a random peptide library (SXXXXXXRG) and I am finding a large difference in IDs between DIA-NN versions 2.3.2 and 2.6.1 (~30K vs ~100). Attached are the logs and runs report for a representative .raw file that I searched with both versions (using predicted spectral library). Troubleshooting with setting the calibration accuracy to 100 ppm did improve the situation. Is there any difference between the versions that could be causing this? We are also aware that the nature of the peptide library results in many isobaric precursors/fragments at the MS1 and MS2 level which may complicate the search.

Thanks in advance,

Oliver

[vdemichev]

Hi Oliver,

You used --dg-keep-cterm 2 for 2.6.1 but not 2.3.2. I would think this option makes sense indeed for this scenario. Also, Proteoforms mode likewise makes sense. I would not trust results without this option.

Best,
Vadim

[Alpine2048]

Hi Vadim,

Thank you for catching that! I am still confused as I have searched the same sample (along with 3 similar samples) in 2.3.2 with --dg-keep-cterm 2 (see attached log) and it yields a comparable amount of IDs (slightly higher) in the first-pass search.

Best,

Oliver

OH090.log.txt

[vdemichev]

Would you please be able to share (email me a link) one run with the respective database, I will then investigate?

[Alpine2048]

Hi Vadim,

I believe I've figured out the problem, which was an oversight on my end. The original search database contained all possible combinations of the random peptide possible, and I believe this may have interfered with the decoy generation/scoring. I recently sequenced the library to narrow down the number of peptides actually present in the library (~10% of total) and used this to search. DIA-NN 2.6.1 is working as intended (~20K IDs). Interestingly, DIA-NN 2.3.2 yields similar amounts of IDs, although when I include entrapment peptides in the database, it suggests that the proteoform scoring does not control the FDR properly at 5% (it is at ~30%) while 2.6.1 does. So I think previously I was getting more IDs in 2.3.2 just because the scoring was lax.