Predicted Gene Expression: High Power, Mechanism, and Direction of Effect



 DESCRIPTION (provided by applicant): Although investments in genomic studies of mental disorders enabled the discovery of thousands of robustly associated variants with these complex diseases, the translation of these discoveries into actionable targets has been hampered by the lack of a mechanistic understanding on how genome variation relates to phenotype. Moreover, it has been widely shown that a substantial portion of the genetic control of complex traits, including mental disorders, is exerted through the regulation of gene expression. However, effective methods to fully harness this mechanism are lagging. To address these challenges, we propose a novel gene-based test -PrediXcan- that directly tests this regulatory mechanism and substantially improves power relative to single variant tests and other gene-based tests. PrediXcan is inherently mechanistic and provides directionality, highlighting its potential utility in identifying novel targets for therapy. The method consists of predicting the whole genome effect on expression traits and correlating this effect with disease risk to identify novel disease genes. In addition, we propose novel approaches to investigate the context-specificity of expression traits (Orthogonal Tissue Decomposition) and to quantify the collective effect of the regulated transcriptome on phenotypes of interest (Regulability). Regulability is similar to the concept of chip heritability (total variability explained collectivey by genotyped variants). First, we will develop cross-tissue, tissue-specific (for over 30 different human tissue types), and brain-region specific expression traits. We will use statistical machine learning methods to develop whole genome prediction models for these traits and extend this work to other molecular phenotypes. All models will be stored in open access databases. Next, we will apply the PrediXcan method to 7 mental disorder phenotypes. More specifically, we will compute genetically predicted levels of gene expression traits and correlate them with disease risk to identify genes involved in disease pathways. We will also quantify the collective effect of the predicted transcriptome (regulability) on mental disorder risk across multiple tissues. Finally we will extend PrediXcan method and develop a method to infer the results of PrediXcan using summary statistics data as opposed to individual level data. This will extend the applicability of the approach to all summary results generated by meta-analysis consortia and increase the power to discover novel genes given the larger sample sizes. The research we propose is driven by an extensive set of preliminary studies, and promises substantial deliverables in both new methods of analysis and public access results databases.