Inference of the human polyadenylation code

Abstract

MotivationProcessing of transcripts at the 3′-end involves cleavage at a polyadenylation site followed by the addition of a poly(A)-tail. By selecting which site is cleaved, the process of alternative polyadenylation enables genes to produce transcript isoforms with different 3′-ends. To facilitate the identification and treatment of disease-causing mutations that affect polyadenylation and to understand the sequence determinants underlying this regulatory process, a computational model that can accurately predict polyadenylation patterns from genomic features is desirable.ResultsPrevious works have focused on identifying candidate polyadenylation sites and classifying tissue-specific sites. By training on how multiple sites in genes are competitively selected for polyadenylation from 3′-end sequencing data, we developed a deep learning model that can predict the tissue-specific strength of a polyadenylation site in the 3′ untranslated region of the human genome given only its genomic sequence. We demonstrate the model’s broad utility on multiple tasks, without any application-specific training. The model can be used to predict which polyadenylation site is more likely to be selected in genes with multiple sites. It can be used to scan the 3′ untranslated region to find candidate polyadenylation sites. It can be used to classify the pathogenicity of variants near annotated polyadenylation sites in ClinVar. It can also be used to anticipate the effect of antisense oligonucleotide experiments to redirect polyadenylation. We provide analysis on how different features affect the model’s predictive performance and a method to identify sensitive regions of the genome at the single-based resolution that can affect polyadenylation regulation.Supplementary information Supplementary data are available at Bioinformatics online.

Sequence determinants of polyadenylation-mediated regulation

“…Previous work applying deep learning approaches to polyadenylation was either applied to endogenous data (Gao et al 2018;Leung et al 2018) or to randomly mutated reporter constructs (Bogard et al 2019). When learning on endogenous data, multiple factors other than the PASs, such as nucleosome composition and epigenetic modification (Lutz and Moreira 2011), can have an effect on polyadenylation.…”

“…Some research efforts employed machine learning and deep learning approaches to prediction of alternative polyadenylation events and classification of sequences as PASs (Cheng et al 2006;Akhtar et al 2010;Chang et al 2011;Gao et al 2018;Leung et al 2018;Bogard et al 2019). The deep learning approaches highlight the usefulness of convolutional neural networks (CNNs) for regulatory genomics and provide valuable predictions for PAS classification and isoform choice.…”

DeepPASTA: deep neural network based polyadenylation site analysis

“…Again, these signals are known as the polyadenylation signals (PASs) in the literature [Hu et al, 2005], [Derti et al, 2012] and [Tian et al, 2005]. Table S2: Performance comparison between DeepPASTA and Conv-Net [Leung et al, 2018] in relatively dominant polyA site prediction on dataset 5 using AUC and AUPRC. [Leung et al, 2018]…”

“…Table S2: Performance comparison between DeepPASTA and Conv-Net [Leung et al, 2018] in relatively dominant polyA site prediction on dataset 5 using AUC and AUPRC. [Leung et al, 2018]…”

Predicting the Impact of cis-Regulatory Variation on Alternative Polyadenylation

“…48 49 Recent work in computational biology, often using deep learning approaches, has taken steps 50 towards building predictive models of regulatory processes. Examples include DeepBind 51 (Alipanahi et al, 2015), Basset (Kelley et al, 2016), DeepSea (Zhou and Troyanskaya, 2015), 52 and even a recent model of APA (Leung et al, 2017). However, the quality of such functional 53 models is determined not only by the underlying algorithm and model architecture but by the 54 quality and size of the training data.…”