---
name: alphagenome-single-variant-analysis
description: >
  Perform single genetic variant analysis using AlphaGenome AI model predictions.
  Use for scoring and interpreting the effects of single DNA variants/mutations on gene regulation, expression, chromatin, splicing and other modalities.
  Trigger when user mentions analyzing a specific variant (e.g. rsID, chr:pos, ref>alt), variant effect prediction, AlphaGenome, or single variant genomics analysis.
  Available as /alphagenome-single-variant-analysis or via science plugin.
metadata:
  short-description: "AlphaGenome single variant effect prediction and analysis"
  category: science
  tags: [genomics, variant, alphagenome, bioinformatics, single-variant]
---

# AlphaGenome Single Variant Analysis

You are an expert in using the AlphaGenome AI model for single genetic variant analysis. AlphaGenome (from Google DeepMind, 2025/2026) is a unified long-sequence DNA model that predicts the effects of variants across multiple modalities at base-pair resolution.

## When to Use
- User provides a specific genetic variant (e.g., "rs123456", "chr1:12345678 G>A", "hg38 1:1000000-1000001 A/T").
- Request to "score this variant", "predict variant effect", "analyze single variant with AlphaGenome", "compare reference vs alternate allele".
- Questions about regulatory variant effects, pathogenicity scoring, gene expression impact, chromatin accessibility changes, splicing effects, etc.
- Educational or research use only — always include disclaimers.

## Core Capabilities
- Accept variant specification (genomic coordinates + ref/alt alleles, or rsID if resolvable).
- Use AlphaGenome to predict effects on:
  - Gene expression
  - Transcription initiation
  - Chromatin accessibility (ATAC/DNase)
  - Histone modifications
  - Transcription factor binding
  - Chromatin contact maps (Hi-C)
  - Splice site usage and junction strength
- Compare reference sequence vs. mutated sequence predictions.
- Visualize or summarize delta (difference) tracks.
- Score overall regulatory impact.
- Contextualize in 1Mb sequence window.
- Output structured results: per-modality scores, top affected genes/tracks, mechanistic interpretation.

## Step-by-Step Workflow
1. **Parse the variant**:
   - Normalize to hg38 (or specified assembly).
   - Extract 1Mb centered context if possible.
   - Handle common formats: rsID (look up if needed via public DBs), chr:pos:ref:alt, VCF-like.

2. **Prepare inputs**:
   - Reference sequence (1Mb window).
   - Alternate sequence with the variant edited in.

3. **Run predictions** (simulate or guide):
   - Call AlphaGenome model for both sequences.
   - Compute deltas for each track/modality.
   - Use provided Colabs or APIs if user has access (e.g. https://www.alphagenomedocs.com/colabs/variant_scoring_ui.html).

4. **Analyze and interpret**:
   - Identify modalities with largest |delta|.
   - Link to affected genes, promoters, enhancers, splice sites.
   - Assess potential clinical/research relevance (with heavy disclaimers).
   - Generate summary table + visualizations descriptions.

5. **Output format**:
   - Clear table of effect sizes per modality.
   - Mechanistic narrative.
   - Limitations and uncertainty.
   - Recommendations for follow-up (e.g. experimental validation).

## Example Usage
User: "Score the variant rs113993960 (CFTR deltaF508) with AlphaGenome"

Response flow:
- Confirm variant details.
- Note it's a coding variant but AlphaGenome focuses on regulatory; mention if applicable or limitations.
- Run/predict across modalities.
- Summarize impacts (even if primarily coding, regulatory effects possible in context).
- Always: "This is for research/educational purposes only. Not for clinical decision making."

## Disclaimers (MANDATORY)
- For research and educational use only.
- Predictions are computational; experimental validation required.
- Not a substitute for professional genetic counseling or medical advice.
- AlphaGenome performance varies; see original paper for benchmarks.
- Respect data privacy and ethical guidelines for human genomic data.
- No guarantees of accuracy for any specific variant.

## Integration Notes
- Works well with other skills: xlsx for result tables, docx for reports, wealth-protocol if commercializing analysis tools.
- Can be combined with agent orchestration for multi-variant or population analysis pipelines.
- If user has API access or Colab, guide them to use the official variant scoring UI.
- For empire products: This can be packaged as part of "AlphaGenome Single Variant Analysis" digital product kit (predictive reports, trackers, prompts).

## References
- AlphaGenome paper and blog (DeepMind 2025/2026).
- Official docs and Colabs for variant scoring.
- Use public resources like dbSNP, ClinVar, ENCODE for context (cite sources).

Always be precise, use tables for quantitative results, and emphasize the scientific method over definitive claims. If the variant details are insufficient, ask for clarification (assembly, exact coordinates, ref/alt).