IndoAutoImmune 2.0HLA Precision Medicine · 5-Stage PipelineHighest-risk module in the stack
HLA-linked hypersensitivity. The one class of AI failure that can kill someone.
HLA-B*15:02 × carbamazepine — 9% of South Asians carry it. SJS/TEN mortality 10–30%. Every US-trained tool assumes the allele frequencies of the population it was trained on. This module rebuilds the pipeline — imputation, novel-allele scoring, Boltz-2 binding, population Monte Carlo, AyurBridge cross-reference — with Indian frequency substrates and a hard benchmark gate that refuses to serve novel-pair predictions unless it correctly flags HLA-B*57:01/abacavir + HLA-B*15:02/carbamazepine as HIGH risk first.
⚠️
LIFE-THREATENING FAILURE MODE · SJS/TEN MORTALITY 10–30%
False negatives on HLA-drug hypersensitivity risk can be fatal. This module ships with the strictest guardrails in the PetriDish stack: novel-pair risk scores are CAPPED below HIGH because a Phase 1 proxy cannot claim confidence a real Boltz-2 co-folding hasn't backed. HIGH tier is emitted only for pairs with FDA / CPIC / PharmGKB literature edges. Never a substitute for laboratory HLA typing before high-risk drug initiation.
✗
PRD §6 BENCHMARK GATE
Loading…
Required: HLA-B*57:01/abacavir AND HLA-B*15:02/carbamazepine flagged HIGH. Failing means
/binding_risk, /population_risk, and /full_pipeline return 503 at request time — a hard architectural guardrail, not a warning.Full 5-stage pipeline · end-to-end
Set at least one HLA tag SNP → pick a candidate drug → optionally add a phytochemical. Pipeline runs impute → novel-allele → binding-risk (benchmark-gated) → population MC → AyurBridge.
TAG SNPS
rs2395029
rs144012689
rs9263726
rs3135388
rs4349859