Paper
Yuan, Durham, Cong, Schaeffer · 2026
TriCyp is the companion deposition for this manuscript. Each main and supplementary figure maps to a navigable surface here so a reader landing from the PDF can re-find the exact panel they want to inspect interactively.
| Figure | What it shows | TriCyp surface |
|---|---|---|
| Fig 1 | Pipeline overview. Pipeline diagram for ESM2-3state classification across ECOD F70 representative domains. | About / Methods → |
| Fig 2 | Held-out benchmark. ROC + PR curves and threshold tuning for ESM2-3state vs SSBONDPredict (disulfide) and vs LMetalSite / GPSite (metal-binding). | Benchmark → |
| Fig 3A | Source-stratified rates. Stacked bars: PDB-geom / PDB-ESM / AFDB-ESM × free-thiol / disulfide / metal-binding fractions. | Dashboard → |
| Fig 3B | Kingdom representation. Domain fraction vs cysteine fraction by superkingdom. | Dashboard → |
| Fig 3C | Per-kingdom rates. Stacked classification rates for Bacteria / Archaea / Eukaryota. | Dashboard → |
| Fig 3D | Subcellular gradient. Eukaryotic subcellular localisation: disulfide and metal-binding rates per compartment. | Dashboard → |
| Fig 4 | AF geometric scanning. Why AlphaFold-monomer geometric scanning is fundamentally limited as a disulfide annotation source — panels A–F with downloadable PyMOL sessions. | AlphaFold geometric scanning → |
| Fig 5A,B | H-group confusion matrix. Structurally-known × ESM2-predicted cysteine fractions per H-group; click-through to the H-groups in any cell. | H-group browser → |
| Fig 5C | Novel metal H-group · 3380.1. Side-by-side PDB-source and AFDB-source representatives with ESM2-predicted metal-binding cysteines highlighted. | H-group 3380.1 → |
| Fig 5D | Novel metal H-group · 804.1. Second highlighted candidate-novel metal-binding H-group. | H-group 804.1 → |
| Fig 5E | Novel metal H-group · 3991.1. Third highlighted candidate-novel metal-binding H-group. | H-group 3991.1 → |
| Figure | What it shows | TriCyp surface |
|---|---|---|
| Fig S1 | Iron-only ROC. Metal-type-stratified ROC. The headline iron-only finding (ESM2 0.993 / LMetalSite 0.917 / GPSite 0.877). | Benchmark → |
| Fig S2 | Source-type breakdown. Source-type breakdown (PDB / AFDB / Prodigal / UniParc) × classification fractions. | Dashboard → |
| Fig S3 | Confidence distribution. Distribution of max-class probability across all classified cysteines. | Dashboard → |
Please cite the manuscript when reusing TriCyp data. Data is released under CC-BY 4.0; the predictor source carries its existing license.Templates below contain DOI/URL placeholders until the bioRxiv preprint is assigned — they will fill in automatically.
@article{yuan_tricyp_2026,
title = {Classification of cysteine fates in structure predictions using a protein language model},
author = {Yuan and Durham and Cong and Schaeffer},
year = {2026},
journal = {bioRxiv},
note = {TriCyp companion site: https://tricyp.swmed.edu},
}TY - JOUR TI - Classification of cysteine fates in structure predictions using a protein language model AU - Yuan AU - Durham AU - Cong AU - Schaeffer, Dustin PY - 2026 JO - bioRxiv ER -
paper/figure_data/ CSVs — see the Downloads page.{ success, data, error } envelope.