variant-interpretation-acmg

Variant Interpretation Agent

ID: biomedical.genomics.variant_interpretation Version: 1.0.0 Status: Beta Category: Genomics / Precision Medicine

---

Overview

The Variant Interpretation Agent automates the clinical assessment of genomic variants (SNVs, Indels). It aggregates evidence from population databases, functional prediction scores (including AlphaMissense), and clinical literature to classify variants according to ACMG/AMP guidelines.

Key Capabilities

1. Annotation & Scoring

• VEP / SnpEff Integration: Functional consequences (missense, frameshift, splice).
• Pathogenicity Prediction:
  • AlphaMissense: Structure-based pathogenicity probabilities.
  • REVEL / CADD: Ensemble scores for missense variants.
  • SpliceAI: Deep learning for splicing effects.

2. Evidence Aggregation

• ClinVar: Checks for existing clinical classifications.
• gnomAD: Population allele frequency filtering (filtering out common benign variants).
• Literature Mining: Searches PubMed for variant-phenotype associations.

3. ACMG Classification

• Automates criteria application (e.g., PVS1, PM2, PP3) to suggest a classification:
  • Pathogenic
  • Likely Pathogenic
  • VUS (Variant of Uncertain Significance)
  • Likely Benign
  • Benign

Usage Example

agent = VariantAgent()
report = agent.interpret(variant="chr7:140453136:A:T", gene="BRAF")
print(report.classification) # "Pathogenic (V600E)"

References

• AlphaMissense (DeepMind, Science 2023)
• ACMG Guidelines (Richards et al., 2015)