A deep boosting based approach for capturing the sequence binding preferences of RNA-binding proteins from high-throughput CLIP-seq data

Abstract

Characterizing the binding behaviors of RNA-binding proteins (RBPs) is important for understanding their functional roles in gene expression regulation. However, current high-throughput experimental methods for identifying RBP targets, such as CLIP-seq and RNAcompete, usually suffer from the false negative issue. Here, we develop a deep boosting based machine learning approach, called DeBooster, to accurately model the binding sequence preferences and identify the corresponding binding targets of RBPs from CLIP-seq data. Comprehensive validation tests have shown that DeBooster can outperform other state-of-the-art approaches in RBP target prediction. In addition, we have demonstrated that DeBooster may provide new insights into understanding the regulatory functions of RBPs, including the binding effects of the RNA helicase MOV10 on mRNA degradation, the potentially different ADAR1 binding behaviors related to its editing activity, as well as the antagonizing effect of RBP binding on miRNA repression. Moreover, DeBooster may provide an effective index to investigate the effect of pathogenic mutations in RBP binding sites, especially those related to splicing events. We expect that DeBooster will be widely applied to analyze large-scale CLIP-seq experimental data and can provide a practically useful tool for novel biological discoveries in understanding the regulatory mechanisms of RBPs. The source code of DeBooster can be downloaded from http://github.com/dongfanghong/deepboost.

Prediction of mRNA subcellular localization using deep recurrent neural networks

“…The goal of RNATracker is to predict an mRNA’s subcellular localization profile from its sequence alone (including possibly its secondary structure inferred from the sequence). To this end, we designed a convolutional bidirectional LSTM neural network with attention mechanism, inspired from previous work on the prediction of protein–mRNA interactions (Alipanahi et al , 2015; Li et al , 2017; Pan and Shen, 2017) and DNA function (Quang and Xie, 2016). Here, we introduce the methodological aspects of training data, feature encoding, model architecture, training and evaluation.…”

“…We introduce, evaluate and interpret RNATracker, a deep neural network predictor of subcellular localization combining two convolutional layers, a bidirectional LSTM layer and an attention module. Although the architecture of our model has some similarities with previously proposed approaches (Li et al , 2017; Pan and Shen, 2017; Quang and Xie, 2016), mRNA subcellular localization differs from most previous applications of deep learning to biological sequence function prediction in several aspects that make it particularly challenging. First, the process of subcellular localization is a long chain of complex events mediated by a large number of protein–RNA and RNA–RNA interactions, and may depend on both primary sequence and secondary structure.…”

“…long short-term memory (LSTM) (Hochreiter and Schmidhuber, 1997)]. These approaches were shown to be highly effective at deciphering complex regulatory mechanisms, such as alternative splicing (Leung et al , 2014), transcriptional regulation (Alipanahi et al , 2015; Quang and Xie, 2016; Zhou and Troyanskaya, 2015), RBP binding (Li et al , 2017; Pan and Shen, 2017) and RNA polyadenylation (Delong et al , 2018). In those approaches, feature extraction and learning are combined in an end-to-end fashion that often yields better performance compared to conventional feature engineering approaches.…”

PRIME-3D2D is a 3D2D model to predict binding sites of protein–RNA interaction

“…Recently, many teams are working to predict protein–RNA-binding sites. Current methods for predicting protein–RNA interaction binding sites are divided into two major categories: predicting the RNA-binding sites on proteins 31 – 45 , and predicting the protein-binding sites on RNA 46 – 51 . These methods usually consider the sequence, structure or physicochemical characteristics of the given protein or RNA.…”

“…With the development of high-throughput sequencing methods, thousands of protein-binding sites on RNA have been discovered. Protein-binding site predictors based on sequencing data were also developed 46 – 51 . However, these methods currently train the model mainly for specific proteins, so they are not universal.…”