Protein Inference - Usage Guide

Overview

Protein inference decides which proteins are present from identified peptides. It is fundamentally underdetermined: bottom-up MS sees peptides, many peptides map to multiple proteins, so the protein set is a chosen explanation under an assumption (parsimony or a probability model), not a measurement. The honest reporting unit is a protein GROUP -- proteins indistinguishable by the observed peptides -- with a designated leading protein, not a flat list. This skill covers grouping, inference-method choice, and protein/group-level FDR done correctly (picked FDR, no two-peptide rule).

Prerequisites

pip install pyopenms pandas
# CLI alternatives: ProteinProphet (TPP), Epifany (OpenMS), Philosopher (FragPipe)

Quick Start

Tell your AI agent what you want to do: - "Group these proteins by shared peptide evidence and give each group a leading protein" - "Apply parsimony to find the minimal protein set explaining all my peptides" - "Run EPIFANY for probabilistic inference with protein-group FDR" - "Estimate protein-group FDR with picked FDR, not the PSM formula" - "Quantify on unique peptides only to avoid razor-peptide artifacts"

Example Prompts

Protein Grouping

"Parse MaxQuant proteinGroups.txt and explain the protein group structure and leading proteins"

"Identify protein groups whose proteins share all peptide evidence (indistinguishable)"

"List proteins with unique peptides versus those identified only by shared peptides"

Inference Methods

"Apply the parsimony principle to report the minimal protein set explaining all peptides"

"Run EPIFANY for Bayesian protein inference with protein-group FDR control"

"Use ProteinProphet to apportion shared peptides across candidate proteins"

FDR Control

"Estimate protein-group FDR with picked-group FDR and report groups at 1%"

"Explain why my 1% PSM FDR left a much higher protein FDR on this deep dataset"

"Flag single-peptide identifications and score them individually instead of dropping them"

Reporting

"Build a protein-group table with the leading accession, unique-peptide count, and group members"

"Summarize how many protein groups pass at 1% picked FDR"

What the Agent Will Do

1. Load the FDR-filtered peptide-to-protein evidence (idXML, proteinGroups.txt, or a PSM table)
2. Build protein groups, collapsing indistinguishable proteins and dropping subsumable ones
3. Apply the chosen inference method (parsimony or probabilistic) and assign a leading protein per group
4. Estimate protein/protein-group FDR with picked FDR, not the reused PSM formula
5. Report groups at the FDR cutoff, judging single-peptide IDs by score rather than a two-peptide rule
6. Hand the inferred groups to quantification, noting the razor-vs-unique choice

Tips

• Report protein groups with a leading protein, never a flat protein list -- a flat list double-counts indistinguishable proteins and breaks protein-level FDR.
• Protein FDR is a separate estimation problem from peptide FDR and inflates on large data; use picked-group FDR (Savitski 2015; The & Kall 2016).
• Do not impose a two-peptide rule -- it raises protein FDR and discards real single-peptide proteins (Gupta & Pevzner 2009).
• "Unique" is database-relative; a peptide unique against SwissProt may be shared once isoforms and TrEMBL are added. Fix and document the database.
• For sensitive differential abundance, quantify on unique peptides only; razor assignment can flip between conditions and fake DE.
• A protein group is not a proteoform -- do not claim a specific isoform from shared-peptide evidence.

Related Skills

• peptide-identification - Produces the FDR-filtered peptide list that feeds inference and shares the target-decoy machinery
• quantification - Consumes inferred groups; razor-vs-unique peptide choice lives here
• data-import - Loads idXML/mzML identification files
• database-access/uniprot-access - Canonical-vs-isoform databases drive uniqueness and the leading-protein convention