Unified calibration across diverse DNS tools

Determine whether a single unified calibration model can reliably calibrate confidence scores and control false discovery rate (FDR) for predictions generated by multiple, diverse de novo peptide sequencing (DNS) tools, or whether separate tool-specific calibration models are required to achieve accurate calibration and FDR control across tools.

Background

The paper introduces Winnow, a model-agnostic framework that calibrates confidence scores from DNS outputs and estimates FDR using a decoy-free, non-parametric method. The authors propose that calibrated scores can facilitate integration, filtering, and ensembling across DNS models and database search engines in hybrid workflows.

In considering broader applicability, the authors note uncertainty regarding whether a single, unified calibration model can accommodate predictions from multiple DNS tools, or if separate calibrations tailored to each tool are necessary. This question impacts the feasibility of cross-tool integration and standardization of calibrated outputs within proteomics pipelines.