IndoVarClassifier v1.0Grant-research pipeline · real methodsPublished methodology · fully cited
The real-methods pipeline. Every score sourced.
IndoVarClassifier v1.0 is the grant-research core of PetriDish — the pipeline you cite in a BIRAC / DBT / Wellcome application. Published methods, live public APIs, deterministic tier synthesis, and a complete evidence chain on every output. No proxy language, no swap-in disclaimers: this is what the methods section looks like.
The 5-stage pipeline
STAGE 1
Population
Ensembl VEP + gnomAD v4 + ClinVar + IndiGen
STAGE 2
ESM-2 zero-shot
BLOSUM62 + Shannon-entropy conservation (Hopf 2017)
STAGE 3
Graph link prediction
NetworkX PGx graph · CPIC / PharmGKB / IMPPAT
STAGE 4
AyurBridge herb x-ref
IMPPAT 2.0 phytochemical-CYP inhibition edges
STAGE 5
Deterministic synthesis
Rule-based tier assignment · no LLM in clinical path
Run a curated demo case
skip_live_apis=true · IndiGen + graph onlyGrant-application narrative
Reproducibility
Every score is deterministic given the same input. The IndiGen proxy table, curated PGx graph, and BLOSUM62 scorer are versioned and inspectable at /curated_substrate.
Open science
All primary data sources are open (Ensembl, gnomAD, ClinVar, PharmGKB, IMPPAT). Full citations at /citations. No proprietary black-box models.
Novel India signal
The pipeline flags variants present in IndiGen but absent from gnomAD global — the exact hypothesis-generation output a grant reviewer looks for.
Clear IP boundary
Deterministic tier synthesis is unpatentable methodology; the differentiating asset is the curated Indian-frequency knowledge graph.