Generative AI Frameworks for Precision Carrier Screening: Transforming Genetic Testing in Reproductive Health

Abstract

Precision carrier screening was developed as a bioinformatics application to inform patient-specific reproductive risks by simulating the Mendelian diseases that offspring of a couple might develop given parental genotype profiles. It became clear that general, robust methods for personalized genomic risk assessments would impact numerous clinical applications across medical genetics. We present GENCORE, a service for general-purpose generative AI for precision carrier screening and any response to specific genetic testing case needs in diagnostic and clinical applications. GENCORE builds upon previously developed generative simulations of protein sequences and inherited transmission outcomes of genotype combinations, extending them to additional sequence and length input spaces and mammalian variant annotations. The primary, innovative contribution we describe here challenges the traditional evaluation of AI as black-box tests against gold standards. Even excellent generative simulations are well-defined tools, not universal oracles. For domain-expert clinicians exploring large-scale variants in genetic testing to guide gestation decision-making, the significance of the absence of interest in simulations is at least as important as the conservativeness of the drawings produced for perturbations requested.