Percolator

Percolator applies semi-supervised machine learning to re-score peptide-spectrum matches (PSMs) from liquid chromatography-tandem mass spectrometry (LC-MS/MS) shotgun proteomics data to improve peptide identification accuracy.

Key Features:

• Semi-supervised learning: Uses semi-supervised machine learning to leverage both labeled and unlabeled examples during training.
• PSM re-scoring: Re-scores peptide-spectrum matches (PSMs) to distinguish correct matches from decoy matches.
• Iterative refinement: Iteratively refines its model during training to improve discrimination between true and decoy PSMs.
• Training data: Trains on both labeled data (known true positives and negatives) and unlabeled data.
• Data compatibility: Operates on LC-MS/MS shotgun proteomics data including tryptic and non-tryptic digests.
• Performance gains: Reported to increase correct peptide assignments by ~17% on tryptic Saccharomyces cerevisiae datasets and up to ~77% for non-tryptic digests.

Scientific Applications:

• Peptide identification: Improves peptide identification accuracy in shotgun LC-MS/MS proteomics experiments.
• Protein quantification: Increases confidence of peptide identifications used for protein quantification.
• Post-translational modification studies: Enhances identification confidence for post-translational modification (PTM) analyses.
• Biomarker discovery: Supports biomarker discovery by improving reliability of PSMs from complex biological samples.

Methodology:

Employs semi-supervised machine learning trained on labeled and unlabeled PSMs, iteratively refining the model to re-score PSMs and separate correct matches from decoy matches.