False discovery rate control enables trustworthy deep-learning-based de novo peptide sequencing

Counterpart Calibration Theory (CCT) enables conservative false discovery rate estimation for deep-learning de novo peptide sequencing, improving trustworthy thresholding in {pi}-NovoQC across peptide-spectrum match and peptide levels.

• Why it matters: Confidence thresholds in de novo peptide sequencing do not consistently reflect false discovery rates, limiting reliable integration of database-unidentified identifications into proteomic analyses.
• What they did: The authors develop CCT using counterpart counts to estimate incorrect target predictions, implement it in {pi}-NovoQC with {pi}-xNovo models, and evaluate FDR estimation and correct-identification retention across sequencing models and datasets.
• The result: Across evaluations, CCT supports conservative FDR estimation and threshold selection, and in the Deep Human Proteome dataset it recovers low-abundance peptide evidence that extends protein-group inference and quantification at specified FDR thresholds.