Predicting regulatory variants using a dense epigenomic mapped CNN model elucidated the molecular basis of trait-tissue associations

Abstract

Assessing the causal tissues of human complex diseases is important for the prioritization of trait-associated genetic variants. Yet, the biological underpinnings of trait-associated variants are extremely difficult to infer due to statistical noise in genome-wide association studies (GWAS), and because >90% of genetic variants from GWAS are located in non-coding regions. Here, we collected the largest human epigenomic map from ENCODE and Roadmap consortia and implemented a deep-learning-based convolutional neural network (CNN) model to predict the regulatory roles of genetic variants across a comprehensive list of epigenomic modifications. Our model, called DeepFun, was built on DNA accessibility maps, histone modification marks, and transcription factors. DeepFun can systematically assess the impact of non-coding variants in the most functional elements with tissue or cell-type specificity, even for rare variants or de novo mutations. By applying this model, we prioritized trait-associated loci for 51 publicly-available GWAS studies. We demonstrated that CNN-based analyses on dense and high-resolution epigenomic annotations can refine important GWAS associations in order to identify regulatory loci from background signals, which yield novel insights for better understanding the molecular basis of human complex disease. We anticipate our approaches will become routine in GWAS downstream analysis and non-coding variant evaluation.

Regulatory SNPs: Altered Transcription Factor Binding Sites Implicated in Complex Traits and Diseases

“…All these facts considerably decrease the efficacy of the available methods for TFBS recognition, most of which are based on the PWM model, which oversimplifies the mechanisms underlying TF–DNA interaction [ 66 , 68 , 69 , 70 ]. Development of new generation bioinformatics approaches relying on machine learning and neural networks raises the hope for more efficient and accurate recognition of both the TFBSs and rSNPs in the genomes [ 190 , 191 , 192 , 193 , 194 , 195 ].…”

DeepFun: a deep learning sequence-based model to decipher non-coding variant effect in a tissue- and cell type-specific manner

“…This resulted in a median AUPRC increase of 0.502 and 0.321 in DNase-seq and histone mark assays, respectively. More details of CNN model construction and performance for each feature can be found in our recent work ( 26 ).…”

“…For each variant, DeepFun considers its neighboring 1000 bp region for context information, and then predicts the active (accessibility or binding) probability of sequence(s) containing either reference allele or alternative allele, respectively. To evaluate the impact of variant, we implemented the previously defined SNP Activity Difference (SAD) or relative log fold change of odds (log-odds) difference between the two alleles ( 26 ).…”

“…To address these challenges, we introduced DeepFun web server, a deep learning model for functional evaluation (DeepFun) of genetic variants and assessment of their effect in a cellular context at single-base resolution. Based on the increasing availability of epigenetics tracks from ENCODE and Roadmap, we constructed DeepFun models ( 26 ) by integrating 1548 DNase I accessibility, 1536 histone mark and 4795 transcription factor binding profiles. It is important to note that our model does not take any specific variant information into consideration.…”

“…novel mutations in any studies and de novo mutations identified from family-based studies). Base on previous comprehensive evaluation ( 26 ), we implemented two functions in DeepFun webserver to predict the functional impact of any query variants. We demonstrated that DeepFun can not only effectively assess the functional impact of a non-coding variant and its impact in a tissue- and cell type-specific manner, but also visualize potential motifs in the region around the variant.…”