Abstract
Transcription factors (TFs) play a key role in transcriptional regulation by binding to DNA to initiate the transcription of target genes. Techniques such as ChIP-seq and DNase-seq provide a genome-wide map of TF binding sites but do not offer direct evidence that those bindings affect gene expression. Thus, these assays are often followed by TF perturbation experiments to determine functional binding that leads to changes in target gene expression. However, such perturbation experiments are costly and time-consuming, and have a well-known limitation that they cannot distinguish between direct and indirect targets. In this study, we propose to use the naturally occurring perturbation of gene expression by genetic variation captured in population SNP and expression data to determine functional targets from TF binding data. We introduce a computational methodology based on probabilistic graphical models for isolating the perturbation effect of each individual SNP, given a large number of SNPs across genomes perturbing the expression of all genes simultaneously. Our computational approach constructs a gene regulatory network over TFs, their functional targets, and further downstream genes, while at the same time identifying the SNPs perturbing this network. Compared to experimental perturbation, our approach has advantages of identifying direct and indirect targets, and leveraging existing data collected for expression quantitative trait locus mapping, a popular approach for studying the genetic architecture of expression. We apply our approach to determine functional targets from the TF binding data for a lymphoblastoid cell line from the ENCODE Project, using SNP and expression data from the HapMap 3 and 1000 Genomes Project samples. Our results show that from TF binding data, functional target genes can be determined by SNP perturbation of various aspects that impact transcriptional regulation, such as TF concentration and TF-DNA binding affinity.
