64. LEVERAGING TRANS-ANCESTRY GENOMICS TO IMPROVE PRECISION PSYCHIATRY FOR SEVERE MENTAL ILLNESS IN DIVERSE POPULATIONS

Bipolar Disorder (BD), Schizophrenia (SCZ), and Major Depressive Disorder (MDD) - collectively referred to as severe mental illness (SMI) - are disabling psychiatric conditions with complex genetic and environmental causes. Hospital-affiliated biobanks that collect patient genetic and electronic health record (EHR) data, such as the diverse U.S.-based All of Us (AoU) cohort and Penn Medicine Biobank (PMBB), provide an opportunity to examine how SMI coincides with other health conditions and environmental factors. We aim to utilize genomic, EHR, and survey data from AoU and PMBB to identify risk factors for SMI. We will then use these variables to develop an algorithm for SMI prediction that will reduce time to diagnosis and treatment initiation. Polygenic scores (PGS) are a promising tool for SMI prediction due to the large genetic component of these conditions. Within the past two years, the largest multi-ancestry genome-wide association studies (GWAS) of BD and MDD to date were published. Additionally, in early 2025 AoU doubled the number of samples with genomic data. As the first step of our project, we used the results from these GWAS to test the performance of PGS for BD, SCZ, and MDD in the new AoU data. We used PRS-CSx on the ∼414,000 individuals with genomic data in AoU v8 to calculate PGS in African (AFR; N=84,148), Admixed American (AMR; N = 79,106), East Asian (EAS; N = 10,099), European (EUR; N=234,353) Middle Eastern (MID; N = 1,545), and South Asian (SAS; N = 5,579) individuals for BD (AFR=2,470; AMR=1,315; EAS=72; EUR=5,269; MID = 21; SAS = 41), SCZ (AFR=2,052; AMR=676; EAS=39; EUR=1,573; MID = 14; SAS = 24), and MDD (AFR=6,996; AMR=5,479; EAS=374; EUR=22,075; MID = 116; SAS = 234). For each phenotype and ancestry group, we calculated three PGS: a within-ancestry PGS derived from matched ancestry summary statistics; a EUR-PGS derived from EUR summary statistics; and a trans-ancestry PGS derived from matched ancestry and EUR summary statistics. Each PRS was tested for association with its matching phenotype via logistic regression, correcting for age, sex, and the top ten within-ancestry PCs. We found that within-ancestry PGS were significantly predictive for nearly all phenotypes for AFR (OR ranges 1.04-1.18; p value range 3.12e-09 - 1.2e-03), AMR (OR range 1.04-1.09; p value range 3.9e-03 - 3.3e-02), and EUR samples (OR range 1.36-1.91; p value range 0.00 - 1.84e-93). To our knowledge, this is the first time that PGS derived from AMR SMI summary statistics have significantly predicted these phenotypes in AMR samples. The EUR-PGS were significantly predictive for all three phenotypes in all ancestry groups (OR range 1.13-3.65; p value range 0.00 - 2.07e-02), with the only exceptions being BD in EAS and MID samples, and SCZ in SAS samples. To our knowledge, this is the first time that SMI phenotypes have been significantly predicted by PGS in SAS and MID samples, although sample sizes were modest. Finally, trans-ancestry PRS had stronger associations with SMI phenotypes in AFR samples. Our analyses demonstrate the benefit of increasing diversity of patients in genetic studies of SMI. We demonstrate that PGS can modestly predict SMI certain phenotypes in ancestry groups with limited representation in biobanks (i.e. MID and SAS). Our ongoing analyses include use of machine learning algorithms to identify EHR variables that help predict SMI development, and integration of our PGS models into these algorithms to improve precision psychiatry.