iProMix: A decomposition model for studying the function of ACE2 based on bulk proteogenomic data for coronavirus pathogenesis

Both SARS-CoV and SARS-CoV-2 use ACE2 receptors to enter epithelial cells in lung and many other tissues to cause human diseases. Genes and pathways that regulate ACE2 may facilitate/inhibit viral entry and replication, and genes and pathways that are controlled by ACE2 may be perturbed during infection, both affecting disease severity and outcomes. It is critical to understand how genes and pathways are associated with ACE2 in epithelial cells by leveraging proteomic data, but an accurate large-scale proteomic profiling at cellular resolution is not feasible at current stage. Therefore, we propose iProMix, a novel framework that decomposes bulk tissue proteomic data to identify epithelial cell component specific associations between ACE2 and other proteins. Unlike existing decomposition based association analyses, iProMix allows both predictors and outcomes to be impacted by cell type composition of the tissue and accounts for the impacts of decomposition variations and errors on hypothesis tests. It also builds in the functions to improve cell type estimation if estimates from existing literature are unsatisfactory. Simulations demonstrated that iProMix has well-controlled false discovery rate and large power in non-asymptotic settings with both correctly and mis-specified cell-type composition. We applied iProMix to the 110 adjacent normal tissue samples of patients with lung adenocarcinoma from Clinical Proteomic Tumor Analysis Consortium, and identified that interferon and{gamma} pathways were most significantly associated with ACE2 protein abundances in epithelial cells. Interestingly, the associations were sex-specific that the positive associations were only observed in men, while in women the associations were negative.