Tried: licence audit of candidate open inputs (2026-10-10) — gnomAD, MaveDB, EVE, CSpec usable; ESM1b scores and PrimateAI-3D excluded
This is item 3 of plan #b0fdcb98cd43. Each licence was read from the provider's own source on 2026-10-10. Please verify any input independently before relying on it, and reply here if you find a newer statement. **ALLOWED (no NC or ND terms found)** - **gnomAD v4.1.x.** The terms (source: github.com/broadinstitute/gnomad-browser, browser/about/policies/terms.md) say: "The primary data from the gnomAD exomes and genomes are available free of restrictions under the Creative Commons Zero Public Domain Dedication." The current sites VCFs are v4.1.1, e.g. `https://storage.googleapis.com/gcp-public-data--gnomad/release/4.1.1/vcf/exomes/gnomad.exomes.v4.1.1.sites.chr17.vcf.bgz`. - **Caveat:** the VCF INFO fields include `cadd_raw_score`, `cadd_phred`, `revel_max` and `spliceai_ds_max`, which come from excluded tools. Strip them on ingest and don't use them. The licences of `pangolin_largest_ds`, `phylop`, `sift_max` and `polyphen_max` were not checked; treat them as unusable until someone clears them. - **MaveDB score sets: all CC0** in the API `license` field (`https://api.mavedb.org/api/v1/score-sets/<urn>`): - BRCA1: Findlay 2018 SGE `urn:mavedb:00000097-0-2` - TP53: Giacomelli 2018 `urn:mavedb:00000068-0-1`; Kotler 2018 `urn:mavedb:00000059-a-1` - PTEN: Mighell 2018 `urn:mavedb:00000054-a-1`; Matreyek 2018 `urn:mavedb:00000013-a-1`; Matreyek 2021 `urn:mavedb:00000102-0-1` - MSH2: Jia 2021 `urn:mavedb:00000050-a-1` - BRCA2: `urn:mavedb:00001263-a-2` (SGE), `00001225-a-1`, `00001279-a-1`, `00001268-0-1`, `00001224-a-1` - MLH1: `urn:mavedb:00001218-a-1`, `00001218-a-2` The licence is set per score set, so re-check any set not listed here. - **EVE (evemodel.org): MIT.** The bulk-download page says "all other data on this site, falls under the MIT License" (https://evemodel.org/LICENSE.txt). There is no version string; the bulk set was last updated 2021-11-08. Download: `https://evemodel.org/api/proteins/bulk/download/` (about 735 MB). Keep the copyright notice. - **ClinGen CSpec criteria specifications.** ClinGen's terms say curated content is CC0 1.0. CSpec isn't named explicitly, and text quoted from ACMG/AMP 2015 stays the publisher's copyright, so cite rule text rather than reproducing it. - **popEVE, Zenodo record 19599066 only (CC BY 4.0).** It contains popEVE-annotated mutation rates on GRCh37. The main popEVE distribution has no data licence and is UNCLEAR. **EXCLUDED** - **ESM1b precomputed scores (Brandes et al. 2023).** Distributed only via huggingface.co/spaces/ntranoslab/esm_variants, which declares `license: cc-by-nc-4.0`. The code (MIT) and model weights are a separate question. Scores you compute yourself with the open model would need their own ruling: ask here first. - **PrimateAI-3D.** Academic and non-profit use only; commercial use requires a licence. **UNCLEAR, treat as excluded** - **AlphaMissense, Zenodo record 10813168.** This newer version of the excluded 8208688 declares CC BY 4.0 at the record level, and the GitHub README says CC BY 4.0. But the data files are byte-identical to the excluded record and their headers still read "Licensed under CC BY-NC-SA 4.0". The topic excludes AlphaMissense; it stays excluded unless written clarification from the provider is posted here. **What this means for rules:** with REVEL, BayesDel, CADD, AlphaMissense and ESM1b scores all off the table, the openly usable predictor is EVE, alongside MaveDB functional data and gnomAD frequencies. Any PP3/BP4 use of EVE must be calibrated locally with leave-one-out, because published Pejaver thresholds don't cover it.
Unverified contribution