ESM-variants (ESM-1v zero-shot variant scoring)

esm-variants applies Meta FAIR's ESM-1v protein language model (Meier et al., "Language models enable zero-shot prediction of the effects of mutations on protein function", 2021) to zero-shot missense variant-effect scoring on human protein-coding variants, developed by the Ntranos lab. It is deployed as a public tool (huggingface.co/spaces/ntranoslab/esm_variants) and is distinct from the base ESM-2 model already catalogued in this registry (data/models/esm2.json): esm-variants is a purpose-built variant-scoring application layer over the earlier ESM-1v checkpoints, not a standalone foundation model.

Marginal call, stated plainly: this entry is a real, fetchable, distinct variant-effect-prediction tool, but it is a thin wrapper/application over an existing base model (ESM-1v) rather than a standalone genomic or protein foundation model in its own right, so it is catalogued here as a marginal fit rather than a clean one.

Licence note: this repository carries its own LICENSE file, confirmed directly (MIT, copyright Nadav Brandes and Vasilis Ntranos, 2023) -- resolving the licence uncertainty an earlier research pass flagged, rather than relying on the underlying facebookresearch/esm repository's separate MIT licence.

Registry note on model_card_uri: the esm-variants tool itself publishes no separate model card. This points instead to the Hugging Face model card of the underlying ESM-1v checkpoint (facebook/esm1v_t33_650M_UR90S_1) it applies for zero-shot scoring, not a GA4GH Genomic AI Model Card v0.6.0-formatted document (no model has yet published one, since it is a newly drafted specification). The primary source repository for the scoring tool itself remains github.com/ntranoslab/esm-variants.