alphagenome-variant-impact-score

Score and annotate genetic variants with AlphaGenome Variant Impact scores via the Atlas API.

Updated Jan 8, 2026
One-click install
npx skills add https://github.com/arslan9024/White-Caves --skill alphagenome-variant-impact-score-arslan9024
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: alphagenome-variant-impact-score
Source: https://github.com/arslan9024/White-Caves/tree/main/.agents/skills/alphagenome_variant_impact_score
Command: npx skills add https://github.com/arslan9024/White-Caves --skill alphagenome-variant-impact-score-arslan9024

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires absl-py, alphagenome, anndata, numpy, pandas, polars, pyarrow, pysam, python-dotenv, tabulate, and includes scripts (resource) components.

What problem does it solve? Prioritizing which genetic variants are functionally important from large VCF callsets or candidate lists is slow and requires integrating many genomic data modalities. This Skill scores variants with calibrated AlphaGenome Variant Impact (AVI) Phred scores, decomposes predictions into 18 biological feature attributions, and ranks candidates so researchers can focus on the highest-impact variants. ## Core Features & Use Cases - Variant Querying & Scoring: Query single or multiple variants in chr:pos:ref>alt format to get Phred scores, quantiles, top percentiles, and 18-modality feature importances with optional track-level provenance. - VCF/Tabular Annotation: Annotate VCF, CSV, TSV, or Parquet callsets in Ensembl VEP CSQ format, rank top candidates, and export machine-readable results. - Saturation Mutagenesis Scans: Score all possible SNVs across a genomic window (chr:start-end) to find functional hotspots. - GENCODE v46 Gene Annotations: Extract exon boundaries, CDS/UTR segments, and canonical or exon-skipping splice junction coordinates for genes and transcripts. - Use Case: Given a VCF of 1,000 rare variants from a sequencing study, annotate the entire callset, filter to variants with Phred >= 20 (top 1% genome-wide), and inspect whether top hits are driven by splicing, chromatin accessibility, or conservation. ## Quick Start Ask the assistant to query the variant chr9:128225994:G>A with the AlphaGenome AVI skill and report its Phred score, top percentile, and top contributing biological modality with an Atlas deep-link.

Frequently Asked Questions about alphagenome-variant-impact-score

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I score a genetic variant with AlphaGenome?▼

Run the CLI query command with a variant in chr:pos:ref>alt format, for example: uv run scripts/alphagenome_atlas_avi.py query "chr9:128225994:G>A" --format json. You need an ALPHAGENOME_API_KEY configured in your environment.

How do I annotate a VCF file with variant impact scores?▼

Use the annotate subcommand with --input and --output flags, optionally setting --top_k and --min_phred thresholds. The tool writes Ensembl VEP CSQ annotations including AVI_PHRED, AVI_QUANTILE, and AVI_TOP_FEATURE into the output VCF.

What does the AVI Phred score mean?▼

AVI Phred is a calibrated score computed as -10 * log10(1 - quantile). Phred >= 40 indicates the top 0.01% of genome-wide SNVs, Phred >= 20 the top 1%, and Phred < 10 the bottom 90% by predicted functional impact.

Can AlphaGenome AVI scores be used for clinical diagnosis?▼

No. AlphaGenome is strictly a research tool and its terms of service prohibit clinical use. Scores reflect predicted molecular and functional impact only and must not be used for medical advice, diagnosis, or treatment decisions.

What input formats does the variant annotation tool support?▼

The annotate command accepts VCF and VCF.GZ files plus tabular formats including CSV, TSV, and Parquet. Tabular inputs must contain chromosome, position, ref, and alt columns, and outputs can be written as VCF, TSV, CSV, Parquet, or JSON.

Why does the GTF query take a long time on first run?▼

The gtf command ingests a 318 MB GENCODE v46 feather dataset requiring about 4.37 GB of RAM. Cold-start downloads take 5-10 seconds on fast networks but up to 60-90 seconds externally; cached queries then run in 4-6 seconds.