RxPredict

Real-time Drug Response Prediction

A sub-100 ms pharmacogenomic prediction API that infers patient drug response from genetic profile, demographics, and history — engineered for low-latency serving with performance gates in CI.

Methods & contributions

• Pharmacogenomics — real star-allele catalog (CYP2D6, CYP2C19, TPMT, DPYD *2A, UGT1A1 *28, SLCO1B1 *5, HLA-B*57:01) mapped to metabolizer phenotypes.
• Latency engineering — optimized scikit-learn pipelines, Redis caching, and feature hashing to hold a <100 ms p99.
• Resilience — circuit breakers and latency benchmarking baked into the request path.
• Performance CI — GitHub Actions gates that fail the build on latency regressions.

Data flow

request → feature hashing → Redis cache → sklearn model
        → response (<100ms p99)   ↘ Prometheus latency metrics

Technologies