LEVERAGING MULTI-MODALITY DATA TO PREDICT GENE EXPRESSION ACROSS BRAIN REGIONS AND BLOOD IN STRESS- AND AGING-RELATED BRAIN DISORDERS

Gene transcription is a highly orchestrated process driven by the interplay between genetic variation and epigenetic modifications. Transcriptomic imputation has advanced the biological fine mapping of GWAS of complex traits; models of genetically-regulated expression (GReX) are used to conduct the transcriptome-wide association studies (TWAS) and detect gene-trait-associations (GTAs) with tissue- and cell-type resolution. This static approach inherently fails to also account for the dynamic epigenetic and environmental contexts that govern gene expression, resulting in missed GTAs. To address this, we introduce MultiPred, an integrative genomic framework that incorporates epigenetic or environmental factors in transcriptomic imputation. MultiPred uses 4 different data modality combinations as predictors of gene transcription, i.e., MethylGPred (genotypes and DNA methylation (DNAm)), MethylPred (DNAm alone), EnviroGPred (genotypes and environmental variables), EnviroPred (environmental variables alone). We trained models in 3 brain regions (prefrontal cortex, hippocampus and amygdala) and in blood with paired genotype, DNAm and gene expression data (n=298, 302, 301 and 494, respectively) to have a more generalized view across the body tissues and by doing comparison of blood-to-brain we gain some biomarker perspective. The newly trained models were also compared with the traditional SNP-based genetically-regulated expression (GReX) models of the same tissues. MultiPred lead to significant increase in gene expression prediction. The proportion of genes with improved accuracy reached 84% in the mPFC using MethylGPred compared to GReX. MethylPred lead to 14208 reliably predicted genes in the mPFC, with 9660 uniquely predicted. We validated the utility of MultiPred by applying it to paired prefrontal cortex genotype and DNAm data (n=662) from the Religious Orders Study and Memory and Aging Project (ROSMAP) dataset. MethylGPred enabled the discovery of GTAs associated with Alzheimer’s disease (AD) and AD neuropathology that are overlooked by TWAS. Our analysis enabled the prioritization of 552 AD neuropathology GTAs, including those critically involved in dopamine and serotonin neurotransmission. Recognizing the barriers to accessing individual-level molecular data, we further developed S-MultiPred, an extended framework specifically engineered for broader accessibility. S-MultiPred enables the identification of trait-associated genes using only GWAS and/or blood-based epigenome-wide association study (EWAS) summary statistics. Applying S-MultiPred(S-MethylGPred and S-MethylPred) to PGC-PTSD GWAS (n= 1,222,653) and EWAS summary statistics (n=4,871), we respectively discovered 5 novel and known PTSD-associated genes from MethylGPred (AHRR, CLN3, EVA1C, MICU3, PATL2) and MethylPred(AHRR, GUSBP5, NOG, PATL2, ZBTB16) that would be undetected by traditional TWAS analysis. These genes are implicated in immune dysregulation and stress granule formation, providing insights into the molecular basis of trauma-related phenotypes. In conclusion, the MultiPred suite represents a significant leap forward in integrative genomics by moving transcriptomic imputation from a static genetic view toward a dynamic, multi-modal view and, thus, providing a scalable, powerful toolkit for unlocking the complex regulatory architecture of human disease.