Group-walk

Group-walk controls the false discovery rate (FDR) in hypothesis testing by incorporating group structure into target-decoy competition frameworks to improve discovery balance in applications such as tandem mass spectrometry-based proteomics.

Key Features:

• Group-aware FDR control: Controls FDR in datasets with inherent group structures where score distributions or proportions of true nulls vary between groups.
• Extension of target-decoy competition (TDC): Addresses limitations of traditional TDC methods that directly compare observed target scores to decoy (knockoff) scores when applied to heterogeneous data.
• Prevents imbalanced discoveries: Incorporates group structure to avoid disproportionate false discoveries in some groups while maintaining overall control.
• Comparison mechanism: Retains the fundamental comparison between target scores and decoy (knockoff) scores central to TDC-based approaches.
• Derived from AdaPT: Builds on the AdaPT framework for controlling FDR with side-information to integrate group-level information.
• Empirical validation: Demonstrates consistent power gains in both simulations and real datasets.
• Quantified performance gains: Reports peptide identification increases such as 4% for precursor charge state at a 1% FDR threshold, 3.6% for peptide length, and 26% for mass differences due to modifications.

Scientific Applications:

• Tandem mass spectrometry proteomics: Improves peptide identification and FDR control in tandem mass spectrometry–based proteomics analyses.
• Grouped hypothesis testing: Applies to any hypothesis-testing scenario with heterogeneous or grouped data where side-information or group structure affects score distributions.

Methodology:

Derived from the AdaPT framework, Group-walk integrates group structure into FDR control by applying target-decoy competition that compares observed target scores to decoy (knockoff) scores, with performance evaluated on simulations and real datasets.