CloudGenomics

Genomic Variant Classification Service

A HIPAA-aware ML service that classifies genetic variants (SNPs, indels) along the ACMG/AMP five-tier spectrum from benign to pathogenic, using population frequency, conservation, and in-silico functional scores.

Methods & contributions

• ACMG/AMP evidence framework — predictions are explained with the standard criteria codes (BA1/BS1 for common variants, PM2 for rarity, PVS1 for loss-of-function, PP3/BP4 for in-silico concordance).
• Calibrated in-silico thresholds — SIFT < 0.05, PolyPhen-2 > 0.85, CADD > 25, REVEL > 0.7, matching published damaging cut-offs.
• VCF processing — parses standard VCF, derives conservation composites (PhyloP, GERP), and flags frameshift vs in-frame indels.
• Compliance posture — VPC endpoints, KMS encryption, WAF, and Step Functions orchestration.

Data flow

VCF → variant parser → feature vector (AF · conservation · in-silico)
     → RandomForest classifier → ACMG tier + evidence codes

Technologies