AI-powered pharmacogenomic inference that predicts adverse drug reactions before the first dose — for every patient, regardless of prior genetic testing.
AI-powered pharmacogenomic inference that predicts adverse drug reactions before the first dose — for every patient, regardless of prior genetic testing.
Adverse drug reactions (ADRs) kill roughly 125,000 Americans every year and injure more than 2 million — a largely preventable toll driven by the same underlying cause: inherited variation in drug-metabolizing enzymes.
Despite decades of pharmacogenomic science and robust clinical guidelines from CPIC3 and ClinPGx4, the overwhelming majority of patients never receive PGx testing — and every existing PGx tool requires a completed genetic test as a prerequisite. The gap is not scientific; it is infrastructural. PGx knowledge exists. The tools to deploy it at the bedside do not.
EHR records, clinical notes, pharmacy histories, and genomic sequences — when available — contain rich, untapped signals of PGx risk. GenoRxAI is building the first AI system that can infer actionable PGx risk from these existing data streams, at population scale, without requiring a prior genetic test.
In our published proof of concept5, a whole-genome sequencing study of 210 Qatari ICU patients found that 91% carried at least one deleterious PGx variant that could affect a drug they were actively receiving — with significant population specificity compared with European and African cohorts6. The signal is real. It is everywhere. It just needs the right model to find it.
GenoRxAI's founders have spent six years developing this approach. The seed fundraise supports Google Cloud compute on the All of Us dataset — 340,000+ individuals with linked whole-genome sequencing and EHR data — to train and validate the models that power PIP, the Pharmacogenomic Inference Platform.
Quantify incidence and prevalence of deleterious variants across 50+ drug-metabolism genes in the full All of Us WGS cohort — stratified by ancestry, age, and sex — and measure overlap with documented ADR events to establish the population-level burden of undetected PGx risk.
Train transformer-based models that predict drug-metabolism phenotype from genotype (G→P) and probabilistically infer variant status from clinical observations alone (P→G) — outperforming rule-based approaches and removing the prior-genetic-test bottleneck.
Release PIP v1.0 as open-source (Apache 2.0, GitHub/Hugging Face), build a FHIR-compliant EHR integration layer, and prospectively pilot at Weill Cornell Medicine / NewYork-Presbyterian — measuring clinician acceptance, alert performance, and time-to-action.
Fewer ADR-related hospitalizations. More accurate initial drug and dose selection. And — for the first time — a way to operationalize pharmacogenomics across real-world healthcare systems regardless of whether a patient has ever had a genetic test.
Anyone who is ever prescribed medication stands to benefit. That is effectively every person, at some point in life.
Bioinformatics data scientist trained at MIT, UCSD, and Weill Cornell. 16 years of experience — 10 in academia, 6 in industry — discovering the role of human genetic variation in disease and developing technology for diagnosis and treatment.
Assistant Professor and computational biologist at Weill Cornell Medicine. PhD in Chemical and Biological Engineering from Tufts University; 9 years of experience at the intersection of genomics, AI, and precision medicine — focused on why individual patients respond so differently to the same drug.
Juan and Roofya are world-class scientists — but building a company takes more than science. GenoRxAI is actively looking for C-level executives to lead business, legal/IP, finance, & HR. If that's you, or if you're an investor looking at seed-stage genomics-and-AI-in-health, we want to talk.
Get in touch