0412 Causal inference of molecular networks integrating multi-omics data

Abstract

Recent developments of massively parallel technologies allow assaying different biological molecules at very high throughput rates, including sequencing and genotyping of DNA, quantifying whole-genome gene expression, including measuring mRNA and microRNA abundance, identifying genome-wide epigenetic modifications, such as DNA methylation, and measuring different proteins and cellular metabolites. These advancements provide unprecedented opportunities to uncover the genetic architecture underlying phenotypic variation. In this context, the main challenge is to decipher the flow of biological information that lies between the genotypes and the phenotypes under study; in other words, the new challenge is to integrate multiple sources of molecular information, i.e., multiple layers of omics data, to reveal the causal biological networks that underlie complex traits. It is important to note that knowledge regarding causal relationships among genes and phenotypes can be used to predict the behavior of complex systems, as well as to optimize management practices and selection strategies. Here, we describe a multistep procedure for inferring causal gene-phenotype networks underlying complex phenotypes integrating multi-omics data. We initially assess marginal associations between genotypes and either intermediate phenotypes (such as gene expression) and endpoint phenotypes (such as carcass fat deposition and muscularity), and then, in those genomic regions where multiple significant hits co-localize, we attempt to reconstruct molecular networks using causal structural learning algorithms. These algorithms attempt to infer networks assuming that the pattern of conditional independencies observed in the joint probability distribution of these set of correlated variables are compatible with the unknown causal model. As a proof of principle of the significance of this integrative approach, we show the construction of causal molecular networks underlying economically relevant meat quality traits in pigs using multi-omics data obtained from an F2 Duroc × Pietrain resource population. Interestingly, our findings shed light on the mechanisms underlying some known antagonist relationships between important phenotypes, for instance, carcass fat deposition and meat lean content. More generally, the proposed methodology allows further learning regarding phenotypic and molecular causal structures underlying complex traits in farm species.