Causal Inference of Disease-Relevant Regulatory Networks

Yale University, New Haven CT

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The primary objective of this proposal is to develop novel statistical methods to identify specific disease- relevant regulatory networks. Such network/systems biology based methods are essential for understanding the role of genes in complex human disease, and have a broad application in areas such as cellular function underlying the disease, drug development, and disease classification. Based on our published works on Graphical Models, we will develop such systems-biology based statistical methods that combine genotype, gene expression, and clinical phenotype data to establish causal disease-related gene networks, through the following three interrelated Specific Aims. Each Aim will focus on a different biological problem. In Aim 1, we propose a novel statistical framework using naturally occurring DNA polymorphisms as randomized perturbations in lieu of experimental perturbation, to effectively infer the causality within the networks for better understanding for network dynamics. First, we perform a global search of disease-relevant epistasis network. Second, we zoom in and focus on the partial networks or modules related to genes of interest based on literatures or previous GWAS results. At next stage we introduce the effect of genotypes as randomized perturbation and establish the relevant eQTL and their effects on the modules. Finally, if the SNP effects are strong enough we can use the information to orient the edges within the network and infer causality and directionality. We will apply our method on two asthma cohorts to identify specific regulatory networks associated with asthma susceptibility and exacerbation. For Aim 2 we will develop novel methods for association between regulatory networks and disease subtypes based on phenotypic networks. First, we will develop a model for building a phenotypic network with multiple correlated disease phenotypes. Next we will tests for association between regulatory networks and different patterns of correlations among phenotypes. Finally, we will use the information from the regulatory networks and their correlations with different phenotypic network patterns for identification of different disease subtypes. We will apply the method on the COPDGene data set to identify regulatory networks associated with different COPD subtypes and phenotype patterns. Aim 3 will be focus on development of novel methods for network pharmacology, in particular, identification of specific networks/modules associated with drug use and its effects on patient outcomes. For that purpose we will consider drug use and treatment plan as pertubations, and develop a model that describes the causal relationship between regulatory networks, treatments and clinical phenotypes. We will apply the method on gene expression data from immortalized B-cells treated with Dexamethasone or sham to identify specific networks/modules associated with inhaled corticosteroids (ICS) use and their effects on asthma patient outcomes. We believe the proposed method will give more insight to the underlying biology of differential drug response and lead to development of more effective treatment plan for patients.