henhao
Sign in
Molecular BiologyJournal of Proteome Research
Peer-reviewed journal article

Beyond Delta Masses: MS Andrea Directly Resolves Combinatorial Peptide Modifications in Open Searches

Louise Marie BuuriD, Stephan WinkleriD, Viktoria DorferiD

PDFDOIPreprintCodeData
Post to X
Threads on XPost to BlueskyThreads on Bluesky
In-depth full-text digest available · Jump to digest ↓

Abstract

Open modification search (OMS) strategies have gained popularity in mass spectrometry-based proteomics for identification of peptides carrying unknown or unexpected post-translational modifications. However, most OMS engines report only the overall mass difference between precursor and matched peptide, without explicitly identifying or scoring combinations of modifications at the PSM level. Here, we introduce MS Andrea, a novel OMS search engine that directly identifies and scores combinations of modifications without predefining them. MS Andrea uses a sequence tag-based strategy to filter candidate peptides, evaluated using the MS Amanda scoring function. First fixed modifications only, then combinations of modifications from the Unimod database based on the observed mass shift. We evaluated MS Andrea using a human histone dataset and two phosphopeptide datasets (HeLa cells and Arabidopsis thaliana), comparing its performance with MSFragger and Sage. Across datasets, MS Andrea identified the highest number of PSMs at 1 % FDR using the standard target-decoy approach while achieving higher or comparable numbers using model-based FDR estimation. Importantly, MS Andrea reports modification identities and sites for up to four modifications at the PSM level. Together, these results demonstrate that MS Andrea enables more detailed, interpretable characterization of peptide modifications while maintaining competitive identification performance in OMS-based proteomics.

02 /Specification & Classification

Click any tag to browse matching papers
Citations
0
FWCI
0.00
Field percentile
14th percentile
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 4 (22 Sept 2026) · CC BY 4.0

MS Andrea, a tag-filtered open search engine, scores and localizes up to four Unimod modifications per peptide-spectrum match, with identification counts competitive with MSFragger and Sage.

Most open modification search engines, including MSFragger and Sage, report only the net precursor delta mass and leave the underlying modifications to downstream inference, while hybrid tools such as Ionbot explicitly model at most two or three per peptide. MS Andrea targets heavily modified peptides such as histone tail fragments and phosphopeptides by scoring and localizing up to four Unimod modifications directly within each peptide-spectrum match.

Key results

  1. SuggestedMS Andrea turns precursor delta masses into named, site-localized modification combinations, separating multi-modification shifts that delta-mass engines collapse into one modification or leave unexplained.

    First Arabidopsis replicate, top PSMs: scans 33168 and 33407 got two separate phosphorylations where PTM-Shepherd gave one pyrophosphorylation; scan 28757 was a missed-cleavage peptide with phosphothreonine versus an unannotated 208.0636 Da shift. In selected HeLa and histone PSMs with three or four modifications, Sage and MSFragger often gave decoys or different peptides. · MS Andrea identifies modifications on PSM level; Tables

  2. ShownUnder target-decoy FDR, MS Andrea returned the most PSMs on all three benchmarks; with model-based FDR it led on histones but trailed MSFragger with PeptideProphet on HeLa and Arabidopsis.

    Histone, 1% FDR, mean of three replicates: target-decoy ~540 (MS Andrea) vs 285 (MSFragger) vs 24 (Sage); model-based 1000 (MS Andrea with Percolator) vs 418 (MSFragger with PeptideProphet) vs 677 (Sage q-values). HeLa and Arabidopsis counts are given only in supplementary tables. · MS Andrea identifies high numbers of PSMs; Figure legends

  3. ShownMS Andrea's target-decoy FDR estimates matched ground truth from a synthetic phosphopeptide library and an entrapment-based estimate.

    Three Marx et al. synthetic library files at 1% estimated FDR: true FDR 0.64%, 0.99% and 1.28% (average 0.97%). FDRBench entrapment on one file (1:1 target to entrapment): combined-method FDP 1.05%; lower bound and sample methods 0.52%, which the authors attribute to known underestimation by those methods. · MS Andrea Reliably Controls the False Discovery…

  4. ShownSite localization was comparable to a closed MS Amanda search and rarely degraded when phosphorylation was allowed on any residue. Spectra carrying a modification missing from Unimod were still identified, reported with an unexplained delta mass.

    False localization rate (FLR) 11%, 16% and 0.5% on three synthetic files at 1% estimated FDR, comparable to closed-search MS Amanda; with phosphorylation allowed on all 20 residues, 0.88% of HeLa PSMs at 1% FDR placed it on a false residue. With phosphorylation deleted from Unimod, one synthetic replicate kept comparable PSM counts and true FDR, reporting such spectra as unmodified plus ~79.979 Da. · MS Andrea Reliably Controls the False Discovery…

  5. ShownSequence tag filtering cuts the per-spectrum candidate pool far more than the wide precursor window alone. Identifications stay nearly the same and searches run much faster.

    Second HeLa replicate, human proteome of 2,092,395 peptides: the mass window alone left 286,219 candidates per spectrum on average; adding tags left 22,401 (99.41% below database, 92.17% below mass-only). Disabling tag filtering gave a small PSM gain at 1% FDR but a substantially longer search. · MS Andrea efficiently reduces the search space…

Methods, data and reproducibility

Spectra are deisotoped and charge-reduced, then two- to four-residue sequence tags are built from a spectrum graph of amino-acid mass edges; candidates containing tags and falling in a wide precursor window are scored by MS Amanda with fixed modifications, and the top ten are rescored with Unimod combinations (up to four) matching the delta mass.

  • HeLa phosphopeptide-enriched LFQphos runs (Bekker-Jensen et al. 2020) searched against the UniProtKB human proteome UP000005640 · human (HeLa cells) · n = six samples · PXD014525
  • Arabidopsis thaliana flower phosphopeptide-enriched raw files from the Mergner et al. 2020 proteome atlas · Arabidopsis thaliana · n = eight raw files · PXD013868
  • Histones from breast cancer patient tissue (PAT-H-MS, Noberini et al. 2016), Arg-C digest, searched against human H3 and H4 sequences · human · n = three raw files · PXD002669
  • Synthetic unmodified and phosphorylated peptide library (Marx et al. 2013) used for ground-truth FDR, FLR, entrapment and missing-Unimod tests · synthetic peptides based on human phosphopeptides · n = three libraries (full library: 96 seed peptides) · PXD000138
  • HeLa peptides (Dorfer et al. 2018) used for development and testing, not benchmarking · human (HeLa cells) · PXD007750
  • Phosphopeptide-enriched HT29 cell data (Piersma et al. 2015) used for development and testing, not benchmarking · human (HT29 cell line) · PXD001550
  • Unimod modification database, filtered to post-translational entries to build the default modification-combination dictionary · n = 1358 entries; 133 used by default
Methods & tools
MS Andrea v1.0, MS Amanda (scoring function and closed-search comparison), MSFragger v4.4.1 (FragPipe v24.0), PeptideProphet, PTM-Shepherd v3.0.11, Sage v0.14.7, Percolator v3.06.1, FDRBench v1.1.1 (entrapment), ThermoRawFileParser v1.4.5, Unimod, cRAP contaminant database, UniProtKB
Code
https://github.com/hgb-bin-proteomics/MSAndrea
Data & accessions
https://doi.org/10.5281/zenodo.21902795https://www.ebi.ac.uk/pride/archive/projects/PXD014525https://www.ebi.ac.uk/pride/archive/projects/PXD013868https://www.ebi.ac.uk/pride/archive/projects/PXD002669https://www.ebi.ac.uk/pride/archive/projects/PXD000138
Software & environment
MS Andrea v1.0 in C#, CLI for Windows, Linux and macOS (input .mgf/.mzml, output .csv); MSFragger v4.4.1 in FragPipe v24.0 with PTM-Shepherd v3.0.11; Sage v0.14.7; Percolator v3.06.1; FDRBench v1.1.1; ThermoRawFileParser v1.4.5
Reproducibility gaps
Custom R and Python post-processing scripts, including the pin-file converter for Percolator, are not linked; search-time and HeLa/Arabidopsis counts appear only in supplementary tables, whose link is a placeholder ('[INSERT LINK]').
Limitations (6)
  • By default only 133 post-translational Unimod entries are combined; adding chemical-derivative or artefact classes enlarges the search and slows it, and shifts with no Unimod match are reported as the unmodified peptide plus a delta mass, unlocalized. · stated by the authors
  • Localization is imperfect: false localization rates were 11% and 16% on two of three synthetic files, and in the all-residue HeLa phospho search most misplaced sites sat next to a valid site, likely from missing site-determining ions; ptmRS is suggested for refinement. · stated by the authors
  • Default edge tolerances (0.015 Da single, 0.004 Da double) were tuned on Orbitrap data; timsTOF, ZenoTOF and Astral users must widen them manually (e.g., 20-40 ppm for TOF, 20 ppm for Astral). · stated by the authors
  • Cross-engine agreement on the first histone replicate was modest (9.6%, 156 PSMs shared by all three), with part of the disagreement traced to one engine calling a target where another called a decoy. · stated by the authors
  • Benchmarks are small (three histone, six HeLa, eight Arabidopsis files) and the three- and four-modification PSMs are hand-picked examples; ground truth for FDR and localization exists only for phosphopeptides, not for combinatorial histone marks. · noted in this summary
  • Model-based comparisons pair each engine with a different rescoring layer (Percolator, PeptideProphet, Sage internal q-values), so count differences mix search-engine and post-processor effects. · noted in this summary

Who should read:Read this if you run open modification searches on phosphoproteomes or histone samples and want each PSM annotated with specific modifications and sites rather than a bare delta mass.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLaguillo-Gó '23Ramírez '26Polasky '23Bittremieux '18Bittremieux '19Karlic '26
73% semantic matchBioinformatics
Andrea Laguillo-Gómez, Enrique Calvo, Noa Beatriz Martin-Cofreces+3 more·Journal of Proteomics·Jul 16, 2023· 4 citations