PGxAI says Andromeda beats frontier AI models on prescribing guidance

PGxAI says a peer-reviewed study in npj Digital Medicine found its Andromeda system outperformed GPT-5, Claude Opus 4 and Grok on pharmacogenomic prescribing recommendations. The Palo Alto company says the result strengthens the case for AI-guided, evidence-linked medication decisions as U.S. policy and reimbursement shift toward precision prescribing.

Why it matters: - PGxAI is positioning Andromeda as a clinical decision-support tool for safer, more personalized prescribing. - The company says the platform could help clinicians apply pharmacogenomic evidence at scale, a bottleneck that has slowed broader adoption of precision medicine. - The timing aligns with growing U.S. attention on pharmacogenomic testing and deprescribing.

What happened: - PGxAI announced a peer-reviewed study in npj Digital Medicine showing Andromeda outperformed GPT-5, Claude Opus 4 and Grok on pharmacogenomic prescribing recommendations. - The company says the study showed the PGxAI-Recommender received the highest expert rating in a blinded feasibility evaluation. - The evaluation found the system performed better than all tested large language models on clinical accuracy and adherence to CPIC prescribing guidelines, with adjusted p<0.01 versus all models. - PGxAI said it has raised $3 million in seed funding from ATEM Capital, Inkberry and Najashi Holding.

The details: - Andromeda is described as a HIPAA-compliant and SOC 2-compliant clinical decision-support platform. - The system combines pharmacogenomics, drug-drug interactions, drug-condition interactions and real-world evidence in one clinician-reviewable layer. - Andromeda translates genetic variants, co-medications and patient conditions into traceable prescribing guidance at the point of care. - Every recommendation links to its underlying evidence sources. - Licensed healthcare professionals keep final treatment decisions. - The system is designed to identify up to 573 gene-drug pairs. - Andromeda can retrieve nearly 20,000 relevant pharmacogenomic publications. - PGxAI said an agentic AI system continuously updates the evidence base by retrieving and structuring newly published literature. - The company says the scientific foundation of Andromeda was validated in the 2026 study. - PGxAI’s CEO and co-founder Mike Zack said medication response depends on more than a single gene and that Andromeda is designed to synthesize that complexity into evidence-linked guidance.

Between the lines: - The study result is as much about workflow as model power: PGxAI is arguing that clinical utility depends on traceability, governance and reviewability, not just benchmark scores. - PGxAI President and Chief Business Officer Burns C. Blaxall said benchmark performance is only the first step and that implementation and accountability determine clinical value. - PGxAI is also linking Andromeda to a broader policy moment, including nationwide Medicare coverage for certain pharmacogenomic tests by July 2025 and an HHS action plan in May 2026 aimed at curbing psychiatric overprescribing. - The company says established CPIC guidance recognizes that variants in CYP2D6, CYP2C19 and CYP2B6 can affect the metabolism, dosing, efficacy and tolerability of multiple serotonin reuptake inhibitor antidepressants. - Allan Gobbs, executive chairman and co-founder, said deprescribing should be treated as the stop phase of precision prescribing and that AI can help clinicians review and trust the evidence.

What's next: - PGxAI is betting that publication in a peer-reviewed journal will help move Andromeda from benchmark claims toward clinical adoption. - The company’s next hurdle is proving that the platform can integrate into real-world prescribing workflows and deliver measurable patient impact. - Broader uptake will likely depend on clinician trust, reimbursement coverage and health system governance. - PGxAI points to continuous evidence updates and evidence-linked recommendations as the foundation for that rollout.

The bottom line: - PGxAI is trying to make a clear case that specialized, evidence-grounded AI can outperform general frontier models in one of medicine’s hardest tasks: choosing the right drug and dose for the right patient.