A cysteine reactivity atlas for the Nomura group

Over the years, a chemoproteomics lab runs thousands of experiments that each answer one question: which cysteines did the probe label this time? The atlas brings all of those answers together, so the lab can ask the bigger questions across every experiment at once.

01What it does

• One record. Search-engine output from years of experiments, normalized into a single site-level table with each experiment's context kept.
• Calibrated confidence. Every site carries a confidence tier calibrated against decoys, so "seen once" and "seen everywhere" don't look the same.
• Quality control. Suspect runs and sites are flagged and kept out, not quietly averaged in.
• An explorer. A static web app for browsing any protein's cysteines, with their evidence and predicted structure.
• Models. Machine-learning experiments on top (protein language-model embeddings and structure features) to score sites that have never been observed.

02Built with

Python, DuckDB, decoy-based FDR calibration, ESM-2 and AlphaFold features, GPU jobs on Berkeley's Savio cluster, and a hand-written HTML/CSS/JS explorer.

03Status

In active development as lab-internal infrastructure. The data are unpublished, so there is no public link yet. This page will link the atlas once it can be shared.