Rescoring with MS2Rescore on MetaMorpheus .mZID file shows no PSM improvement

[luffy55512]

Hi,

I’m encountering an issue using MS2Rescore 3.1.5 with .mzID files exported from MetaMorpheus (1.1.2). Despite following the documented procedures, rescoring has no positive impact on PSM identification. There’s no visible separation between target and decoy scores, and the number of PSMs at 1% FDR even drops slightly.

---My Setup

1. PSM input: MetaMorpheus .mzID file
2. Spectrum file: Matched .mzML
3. Mods in MetaMorpheus search: Only Carbamidomethyl[C] Acetyl (Protein N-term) and Oxidation[M]
4. MS2PIP: enabled, model = HCD2021

---MS2Rescore output summary
Progress window shows:
Reading PSMs from PSM file (1/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP1_TL_20211230155755-calib.mzID'...
Reading PSMs from PSM file (2/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP1_TL_20211231164002-calib.mzID'...
Reading PSMs from PSM file (3/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP1_TL_20220107122146-calib.mzID'...
Reading PSMs from PSM file (4/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP3_TL_20211231225031-calib.mzID'...
Reading PSMs from PSM file (5/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP3_TL_20220107183225-calib.mzID'...
Reading PSMs from PSM file (6/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP3_TL-calib.mzID'...
Reading PSMs from PSM file (7/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP5_TL_20211231041904-calib.mzID'...
Reading PSMs from PSM file (8/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP5_TL_20220101050100-calib.mzID'...
Reading PSMs from PSM file (9/9): 'C:/Users/Meizhu/Documents/MS2_prediction/PP5_TL_20220108004253-calib.mzID'...
Removed 0 PSMs with rank >= 10.
Found 108915 PSMs, of which 0.72% are decoys.
Found 99941 identified PSMs with rank <= 1 at 0.01 FDR before rescoring.
Adding basic features to PSMs.
Adding MS²PIP-derived features to PSMs.
Running MS²PIP for PSMs from run (1/9) PP1_TL_20211230155755-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (2/9) PP1_TL_20211231164002-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (3/9) PP1_TL_20220107122146-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (4/9) PP3_TL_20211231225031-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (5/9) PP3_TL_20220107183225-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (6/9) PP3_TL-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (7/9) PP5_TL_20211231041904-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (8/9) PP5_TL_20220101050100-calib...
Processing spectra and peptides...
Running MS²PIP for PSMs from run (9/9) PP5_TL_20220108004253-calib...
Processing spectra and peptides...
576 out of 21365 peptides could not be mapped. Please check your digest settings.
Attempting to fix constant PEP values by removing decoy PSMs that score higher than the best target PSM.
Removed 1 decoy PSMs.
Removed 8311 PSMs with rank >= 1.
Identified -118 (-0.12%) less PSMs with rank <= 1 at 0.01 FDR after rescoring.
Writing output to C:/Users/Meizhu/Documents/MS2_prediction/rescore/1.psms.tsv...
❌ log: 'C:/Users/Meizhu/Documents/MS2_prediction/rescore/1.log.txt'
Writing report to C:/Users/Meizhu/Documents/MS2_prediction/rescore/1.report.html

Main problem:

1. All q-values and All PEP values are the same across every PSM
2. The flashlfq_output.tsv file is completely empty
3. The report.html file contains no visual plots, such as target-decoy separation or score distributions
4. Although the report.html shows decoys <1%, the original MetaMorpheus .psmtsv file indicates that ~30% of PSMs are decoys

My goal is to perform quantitative proteomics analysis using the PSMs after rescoring. I intend to use the flashlfq file generated by MS2Rescore as input to FlashLFQ for label-free quantification, and compare the quantitative results to those obtained from MetaMorpheus.

However, since the rescoring currently has no impact (no improvement in PSM confidence)

1.full-config.json
1.mokapot.decoy.peptides.txt
1.mokapot.decoy.proteins.txt
1.mokapot.decoy.psms.txt
1.mokapot.flashlfq.txt
1.mokapot.peptides.txt
1.mokapot.proteins.txt

[luffy55512]

I’ve resolved the earlier issue — it was caused by pre-filtering the MetaMorpheus output at FDR < 0.01. After re-running with FDR set to 1.0 and passing the unfiltered results to MS²Rescore, the decoy proportion returned to a normal ~39%, and both PSM and peptide identifications increased.

However, one problem remains: report.html still contains no visual plots, and the only figure (target/decoy) appears as a straight diagonal line.

[RalfG]

Hi @luffy55512,

Happy to hear you could resolve the issue! Indeed, all unfiltered target and decoy PSMs should be present in the input file to MS²Rescore.

The issue with empty charts has been fixed in #242 and will be released soon.

Best,
Ralf