PGxAI Clinical Engine

An explainable clinical decision-support prototype that finished as the first honorable mention at the BIO4AI Hackathon, BelBI 2026. It evaluates the suitability of 18 antidepressants against a patient profile: CYP2D6, CYP2C19 and CYP2B6 genotype, current medications, clinical risk factors, treatment history and patient preferences.

The engine is rule-based rather than learned, which is the point. Every ranking comes with the factors that produced it, so a clinician can disagree with the reasoning rather than with a score.

This is a decision-support prototype and an exploration of explainability in pharmacogenomics. It is not a diagnostic tool and it has not been clinically deployed or validated.

Pharmacogenomic reasoning does not reduce cleanly to a score. Metabolizer status interacts with concurrent medications through phenoconversion, evidence strength varies by gene–drug pair, and two rules can point in opposite directions.

The engine therefore models metabolism, phenoconversion, drug interactions, safety risks and evidence confidence as separate contributions, and keeps provenance for each so the explanation is generated from the same structure that produced the ranking.

• 01Evaluates 18 antidepressants against a structured patient profile
• 02Models CYP2D6, CYP2C19 and CYP2B6 metabolizer status
• 03Phenoconversion from concurrent medications
• 04Drug interaction and safety risk modelling
• 05Evidence confidence attached to each contribution
• 06Transparent ranking with per-factor explanation
• 07Pharmacogenomic data processing pipeline
• 08Rule validation and testing pipelines
• 09Evidence provenance tracking