PMD-FDR

PMD-FDR applies Bayesian post-analysis of precursor mass discrepancy to improve the accuracy of peptide-spectrum matches (PSMs) in mass spectrometry-based shotgun proteomics.

Key Features:

• Precursor Mass Discrepancy (PMD) Utilization: Leverages discrepancies between observed and expected precursor masses to flag potentially incorrect PSMs.
• Bias Identification and Correction: Identifies and corrects PMD biases related to time of acquisition within an LC-MS run, presence of decoy PSMs, and peptide length.
• Bayesian Confidence Measure: Applies a post-analysis Bayesian approach that integrates search scores and PMD to compute per-PSM confidence.
• Versatility Across Proteomics Projects: Evaluated on standard proteomics, proteogenomics (custom genomic-based plus reference databases), and metaproteomics (microbial community with a conglomerate database).
• Performance Metrics: Detects approximately 60–80% of likely incorrect PSMs while losing about 5% of correct PSMs.
• Diagnostic Use for Method Development: Serves as a diagnostic metric to assess the quality of results from different experimental workflows.

Scientific Applications:

• Proteogenomics: Improves confidence in peptide identifications when integrating genomic-derived databases with reference proteomes.
• Metaproteomics: Enhances peptide identification reliability in microbial community proteomic studies using conglomerate databases.
• Protein Quantification: Refines input PSM sets for downstream protein quantification by reducing high-scoring false positives.
• Functional Annotation: Increases reliability of peptide-based functional annotations by filtering likely incorrect PSMs.
• Biomarker Discovery: Improves the quality of candidate peptide evidence used in biomarker identification workflows.
• Method Development and Optimization: Provides a metric for comparing and optimizing experimental workflows and search strategies.

Methodology:

Analyzes precursor mass discrepancies between observed and expected masses, models and corrects biases due to LC-MS acquisition time, decoy PSM prevalence, and peptide length, and applies a Bayesian post-analysis integrating search scores and PMD to produce per-PSM confidence estimates.