NEXT-GENERATION IMPUTATION FOR PSYCHIATRIC GENETICS: A NOVEL 515K DIVERSE REFERENCE PANEL AND SERVICE

Accurate genotype imputation and cross-dataset harmonization of genomic data are essential for identifying the complex genetic underpinnings of mental health disorders. One of the most significant obstacles to the expanding diversity in genetic studies today is the lack of adequate reference data for individuals of non-European ancestries. While the recruitment of participants from diverse populations has improved, the availability of genomic tools to analyze their data is still lacking. Large electronic health records and biobanks, such as the NIH initiatives All of Us (AoU) and Analysis, Visualization, and Informatics Lab-space (AnVIL) have aggregated hundreds thousands of individuals from diverse ancestries. Here, we present a novel reference panel built from over 515,579 individuals from AoU and AnVIL data. In addition to being the largest reference panel to date, this panel prioritizes ancestral variability, encompassing 261,163 samples from non-European ancestries–a representation nearly twice the size of the entire TOPMed reference panel. Specifically, this panel includes 101,982 individuals with African, 90,553 admixed/Latin American, 13,226 East Asian, 9,710 South Asian, 1,065 Middle Eastern/North African, and 44,627 other non-European ancestries. The high coverage (30x) whole genome sequencing data in 414,830 AoU and 100,749 AnVIL samples were harmonized and QCd using Hail. A total of 665,398,839 high-quality variants from autosomes were exported as VCF files, removing the variants with: number alternate alleles greater than 31, average sum of allele depths (AD) less than 12 (proxy for depth; DP), singletons, variants with excessive heterozygosity, mean genotype quality (GQ) under 30, and call rate under 0.9. Exported VCFs were subsequently phased using Beagle 5.5 and shuffled with RESHAPE for increased security of the sensitive data. This resource will significantly improve the accuracy of genotype imputation, particularly for rare variants and underrepresented populations, empowering novel discoveries in psychiatric genetics. The reference panel will be available for imputing genotype array (using Beagle) and low-pass sequencing (using Glimpse) data through Broad Institute of MIT and Harvard. The service is set to become available mid-2025, and will be first accessible as a command-line tool with a forthcoming web-based user interface.