circuitSNPs: Predicting genetic effects using a Neural Network to model regulatory modules of DNase-seq footprints

Abstract

AbstractMotivationIdentifying and characterizing the function of non coding regions in the genome, and the genetic variants disrupting gene regulation, is a challenging question in genetics. Through the use of high throughput experimental assays that provide information about the chromatin state within a cell, coupled with modern computational approaches, much progress has been made towards this goal, yet we still lack a comprehensive characterization of the regulatory grammar. We propose a new method that combines sequence and chromatin accessibility information through a neural network framework with the goal of determining and annotating the effect of genetic variants on regulation of chromatin accessibility and gene transcription. Importantly, our new approach can consider multiple combinations of transcription factors binding at the same location when assessing the functional impact of non-coding genetic variation.ResultsOur method, circuitSNPs, generates predictions describing the functional effect of genetic variants on local chromatin accessibility. Further, we demonstrate that circuitSNPs not only performs better than other variant annotation tools, but also retains the causal motifs / transcription factors that drive the predicted regulatory effect.Contactfluca@wayne.edu, rpique@wayne.eduAvailabilityhttp://github.com/piquelab/circuitSNPs