Basset: Learning the regulatory code of the accessible genome with deep convolutional neural networks

Abstract

The complex language of eukaryotic gene expression remains incompletely understood. Thus, most of the many noncoding variants statistically associated with human disease have unknown mechanism. Here, we address this challenge using an approach based on a recent machine learning advance-deep convolutional neural networks (CNNs). We introduce an open source package Basset (https://github.com/ davek44/Basset) to apply deep CNNs to learn the functional activity of DNA sequences from genomics data. We trained Basset on a compendium of accessible genomic sites mapped in 164 cell types by DNaseI-seq. Basset predictions for the change in accessibility between two variant alleles were far greater for GWAS SNPs that are likely to be causal relative to nearby SNPs in linkage disequilibrium with them. With Basset, a researcher can perform a single sequencing assay in their cell type of interest and simultaneously learn that cell's chromatin accessibility code and annotate every mutation in the genome with its influence on present accessibility and latent potential for accessibility. Thus, Basset offers a powerful computational approach to annotate and interpret the noncoding genome.

Impact of regulatory variation across human iPSCs and differentiated cells

“…In particular, we aimed to identify DNA sequences that could predict cell-type-specific effects of regulatory variants. We investigated the use of machine learning models to predict the chromatin activity of regulatory elements across our three cell types using DNA sequence only (Zhou and Troyanskaya 2015;Hashimoto et al 2016;Kelley et al 2016;Zeng et al 2016). We developed a four-layered neural network architecture, OrbWeaver, to predict cell-type-specific chromatin accessibility of 500-bp windows centered at a regulatory locus ( Fig.…”

A synergistic DNA logic predicts genome-wide chromatin accessibility

“…Although SCM differs from existing methods aimed at binary classification of hypersensitive and nonhypersensitive chromatin, we asked how SCM performance compares to four sequence-based classifiers that use either k-mer based models (gkm-svm, SeqGL) or deep learning based models (deepSEA, Basset) (Ghandi et al 2014;Setty and Leslie 2015;Zhou and Troyanskaya 2015;Kelley et al 2016). Although SCM is designed for quantitation and not binary prediction, SCM performs as well as the four state-of-the-art binary predictive methods on black-box binary prediction of functional genomic regions (Supplemental Fig.…”

SNPDelScore: combining multiple methods to score deleterious effects of noncoding mutations in the human genome

“…These methods include DeepSEA (Zhou and Troyanskaya, 2015), DeepBind (Alipanahi et al, 2015) and Basset (Kelley et al, 2016) that 'deep learn' regulatory sequence code from big genomics data; deltaSVM (Lee et al, 2015) and deSNPs (Huang and Ovcharenko, 2015;Li and Ovcharenko, 2015) that learn sequence features from a single enhancer-associated chromatin profile and consider the k-mer content associated with the genetic variant only; CATO (Maurano et al, 2015) that predicts chromatin states by using high-throughput sequencing data across multiple individuals; C-SCORE (Kircher et al, 2014) that integrates multiple annotations into one metric by contrasting variants that survived natural selection with simulated mutations; LINSIGHT (Huang et al, 2017) that predict the likelihood of deleterious fitness consequences of mutations at noncoding nucleotide sites by combining a generalized linear model for functional genomic data with a probabilistic model of molecular evolution; and CAPE (Li et al, 2016) that decomposes the sequence code of potential-binding sites and the binding sites of cofactors from a set of chromatin profiles, and directly quantifies the deactivating effect of a single nucleotide mutation based on the corresponding change in the underlying k-mer profile.…”