AI-powered pharmacogenomic decision support for clinicians

🏆 VandyHacks XII Winner | Best Use of AI

Pharmacogenomic (PGx) reports contain valuable genetic insights, but they are often delivered as dense, unstructured documents that are difficult for clinicians to quickly interpret. This creates a gap between available genetic data and real-world clinical decision-making. We were inspired to bridge this gap by building a tool that transforms complex PGx reports into actionable, evidence-based guidance. Our goal was to reduce trial-and-error prescribing and improve patient outcomes through personalized medicine. Genoscript is able to reduce the time and complexity for clinicians when looking at Pgx reports and prescribing medicine.

GenoScript is a clinical decision-support tool that:

• Parses pharmacogenomic (PGx) reports into structured gene-phenotype data plain english
• Stores patient data securely for future reference
• Allows clinicians to input a medication (e.g., sertraline)
• Uses CPIC guidelines to generate evidence-based drug recommendations
• Displays risk levels using intuitive severity flags (red, yellow, green) This enables providers to make faster, more informed prescribing decisions tailored to each patient’s genetic profile.
• Shows other alternative drugs that may be a better option that are effected by the same gene

We built GenoScript using:

• Next.js + TypeScript for the frontend and application logic
• MongoDB for storing patient data and PGx results
• CPIC Pharmacogenomics API to retrieve clinical guidelines
• Gemini to parse unstructured PGx reports into structured gene-phenotype mappings

• Parsing unstructured PGx reports: Reports vary widely in format, requiring robust AI-based extraction prompts to standardize gene-phenotype data
• Mapping to clinical guidelines: Ensuring accurate matching between patient phenotypes and CPIC recommendations required careful data handling
• Balancing usability and clinical accuracy: We needed to present complex medical information in a way that is both precise and easy to interpret
• Time constraints: Integrating multiple systems (AI parsing, database, API lookup, UI) within a hackathon timeframe required prioritization and rapid iteration

• Successfully built an end-to-end pharmacogenomics decision-support tool within a hackathon timeframe
• Transformed unstructured PGx reports into structured, usable gene-phenotype data using AI
• Integrated real clinical guidelines (CPIC) to generate evidence-based drug recommendations and alternatives
• Designed an intuitive, color-coded system (red/yellow/green) to clearly communicate clinical risk
• Created a scalable architecture combining frontend, database, AI parsing, and external APIs
• Bridged the gap between complex genetic data and practical clinical decision-making
• Delivered a polished, functional prototype that demonstrates real-world healthcare impact

• How to integrate AI-driven data extraction into a real clinical workflow
• The importance of standardized medical data structures for interoperability
• How to design user interfaces for healthcare professionals, where clarity and trust are critical
• The potential of combining genomics + AI + decision support systems to improve patient care

• Integration with electronic health records (EHR systems)
• Real-time clinical validation and provider feedback loops
• integrating real patient charts In for hospitals to be able to use the application