Apply novel pathogenomic approaches to identify interpretable image QTLs for multiple normal tissues

ABSTRACT The overarching goal of this project is to identify tissue-specific genetic variants and genes that are associated with cellular and tissue morphology in normal tissues. We plan to apply advanced computational pathology methods based on machine learning and computer vision to analyze normal tissue histology H&E images from Genotype-Tissue Expression Project (GTEx), extracting interpretable image features as quantitative traits. Next, we will apply bioinformatic and statistical genetic methods to identify the morphological traits that are correlated with eQTLs in donor population, thus generate image QTLs (a.k.a. imQTLs) for all normal tissue types in GTEx with over 100 samples. In addition, we will apply advanced functionally informed GWAS (FiGWAS), which has been successfully applied to eQTL research and significantly boosted the detection of rare variants in genomic association study, to further investigate the association of non-eQTL genetic variants with the interpretable quantitative morphology features described above and to generate supplementary imQTLs. Neither of these two approaches has previously been applied to identify imQTLs. The workflow will initially focus on cell type morphological features, then expand to features related to tissue development and cell-cell interactions. The identified imQTLs (or the genes/image traits associated with them) will be further tested in histopathological images in corresponding tissues from The Cancer Genome Atlas (TCGA) for any difference in terms of imQTL presence, abnormality in the associated image traits, or expression in the associated genes. The identified imQTLs will not only generate new insights about the tissue differentiation, development, and morphological variations in the normal population, but also will provide a solid basis for comparing pathological changes in many types of diseases and help quantify the level of the corresponding histopathological changes. The resulted image features and imQTLs will be made available through a web portal called PathoGenome Viewer for general public query and use.