EWCL-Data

Entropy-Weighted Collapse Likelihood (EWCL) benchmark data supporting

Cristino & Uversky, “Entropy-Weighted Collapse Likelihood: Putting Ensemble Thermodynamics First in Disorder Characterization”, Biomolecules (submitted).

This repository provides a landing page, light-weight manifests, and example scripts.
The heavy per-residue CSV/JSON outputs are archived on Zenodo (see below).

---

Archived datasets

1. MobiDB / DisProt, IDEAL, Merge, UniProt & binding snapshots

Zenodo record (primary benchmark archive)
🔗 https://doi.org/10.5281/zenodo.17409146

Contents (per-residue, frozen):

• DisProt, IDEAL, Merge, and curated UniProt references
• Binding-related snapshots (e.g. FuzPred, FuzDrop, AIUPred comparisons)
• EWCLv1 scores and labels used for all tables/figures in the paper

All files here are exactly the ones referenced in the manuscript tables and figure panels.

---

2. CAID-aligned benchmarks (PDB & NOX tracks, linker, binding)

Zenodo record (CAID-aligned archive)
🔗 https://doi.org/10.5281/zenodo.17352525

Contents:

• CAID-aligned disorder, linker, and binding benchmarks
• EWCLv1 per-residue scores and labels on CAID reference sets
• Metrics used for the “CAID-aligned” comparison tables in the manuscript

These data are alignment only and are not a CAID leaderboard submission.

---

3. NMR, X-ray, and AlphaFold overlays

Planned archives (links will be added when the Zenodo records are public):

• NMR rigidity benchmark – EWCL vs per-residue rigidity traces
• X-ray B-factor benchmark – EWCL vs SIFTS-normalized B-factors
• AlphaFold overlays – EWCL-P3 vs pLDDT and hallucination masks

Until those are live, the corresponding figures in the manuscript are generated from a frozen internal snapshot of these datasets.

---

Repository contents

This GitHub repo is intentionally light-weight:

• README.md – this file, with links to all Zenodo archives
• (Optional) manifests/ – text/CSV manifests listing which files are in each Zenodo record
• examples/ – minimal examples showing how to:
  • load the per-residue CSV/JSON files
  • compute AUROC / AUPRC
  • recreate simple EWCL vs DisProt or EWCL vs pLDDT plots

Note: The full benchmark CSV/JSON files themselves are stored on Zenodo,
not committed here, to keep the repository small and versioned cleanly.

---

Frontend / API

The same frozen per-residue outputs are also exposed in read-only form through:

• 🌐 EWCLx frontend: https://ewclx.com

When a given protein is available both in the Zenodo archives and on EWCLx, the overlays and metrics shown on the site are generated directly from the archived per-residue scores.

---

How to cite

If you use these benchmarks or overlays, please cite:

EWCL benchmark datasets (MobiDB/DisProt, IDEAL, Merge, UniProt, binding)

Cristino, L.; Uversky, V.N. (2025). EWCL – MobiDB/DisProt, IDEAL, Merge, UniProt, and Binding Benchmarks (Version 1). Zenodo. https://doi.org/10.5281/zenodo.17409146

EWCL – CAID-aligned benchmarks

Cristino, L.; Uversky, V.N. (2025). EWCL – CAID-aligned Benchmarks (Disorder PDB & NOX, Linker, Binding) (Version 1). Zenodo. https://doi.org/10.5281/zenodo.17352525

BibTeX:

@dataset{ewcl_mobidb_zenodo_2025,
  author    = {Lucas Cristino and Vladimir N. Uversky},
  title     = {EWCL -- MobiDB/DisProt, IDEAL, Merge, UniProt, and Binding Benchmarks},
  year      = {2025},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.17409146},
  url       = {https://doi.org/10.5281/zenodo.17409146}
}

@dataset{ewcl_caid_zenodo_2025,
  author    = {Lucas Cristino and Vladimir N. Uversky},
  title     = {EWCL -- CAID-aligned Benchmarks (Disorder PDB \& NOX, Linker, Binding)},
  year      = {2025},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.17352525},
  url       = {https://doi.org/10.5281/zenodo.17352525}
}