GDM AlphaGenome Variant Analysis skill

Predicts a non-coding genomic variant's effect on gene expression, chromatin accessibility, histone marks, and transcription-factor binding via Google DeepMind's hosted AlphaGenome API.

Unsigned

Overview

The `alphagenome-single-variant-analysis` skill from Google DeepMind's Science Skills collection. Given a variant in chr:pos:ref>alt form (or a gene/region query), it calls the AlphaGenome API to predict effects on RNA-seq expression, DNase chromatin accessibility, ChIP histone marks, and transcription-factor binding, and its own SKILL.md description names the workflow's actual purpose -- 'non-coding variant effects, pathogenicity, clinical significance, disease associations' -- alongside sibling skills in the same repository for `clinvar_database`, `gnomad_database`, and `dbsnp_database` lookups, confirmed live via direct fetch of the skills directory listing. Verified live via direct fetch of the skill's SKILL.md and its `internal project documentation` (a biological-interpretation and pre-report reasoning checklist), plus the repository's published technical report on the Science Skills collection. Terms check: the repository's root LICENSE is Apache-2.0 (confirmed by direct fetch and by `plugin.json`'s `license` field), and the repository's own `SKILL_LICENSES.md` separately documents that individual skills reference third-party data sources (AlphaGenome, ClinVar, gnomAD, etc.) carrying their own terms of use -- a data-usage-terms disclosure, not a code-licence divergence, so Apache-2.0 is recorded as this entry's `license` without qualification. The collection is built and documented primarily for Google's Antigravity harness (distributed via the open Agent Skills standard's `npx skills add`); Antigravity has no matching entry in this registry's closed harness vocabulary (`schemas/vocab/harnesses.json`), so `host_runtimes` is omitted below rather than mapped to an unverified value. No slug collision with any existing entry in this registry.

Details

Licence: Apache-2.0
Version: 1.2.0
Category: Variant interpretation
Homepage: https://github.com/google-deepmind/science-skills/tree/main/skills/alphagenome_single_variant_analysis
Repository: https://github.com/google-deepmind/science-skills
Maintainers: Susheel Varma (@susheel, GA4GH AI Workstream / Sage Bionetworks)
Keywords: alphagenomevariant-effect-predictionregulatory-genomicsgoogle-deepmindclinvargnomad

Safety classification

No safety classification has been submitted for this entry.

Skill

Human oversight: required
Provenance logging: No
Evaluation criteria: https://storage.googleapis.com/deepmind-media/papers/google_deepmind_science_skills_for_antigravity_towards_efficient_and_reliable_scientific_workflows.pdf
Source: https://github.com/google-deepmind/science-skills.git (skills/alphagenome_single_variant_analysis)

Inputs

variant (string) — A genomic variant in chr:pos:ref>alt format, or a gene symbol/genomic region for de-novo annotation.
tissue_or_cell_type (string) — Optional UBERON/CL ontology term scoping the prediction to a tissue or cell-type context.

Outputs

variant_effect_predictions (object) — Predicted effects on gene expression (log TPM), chromatin accessibility, histone marks, splicing, and transcription-factor binding, returned by the hosted AlphaGenome model.

Provenance

Created: 2026-09-08T00:00:00.000Z
Updated: 2026-09-08T00:00:00.000Z
Last verified: 2026-09-08T00:00:00.000Z