RNA-binding proteins (RBPs) play crucial roles in both transcriptional and post-transcriptional gene regulation. RNAs interact with RBPs to conduct their functions and their dysregulation are associated with many diseases. How to computationally detect RNA-binding proteins based on collected data are needed. Previous prediction approaches learns a binary classification model for a target-specific RBP to predict whether a RNA binds to a RBP or not. However, a RNA can bind to multiple RBPs. Thus, when given a RNA sequence, it would be more useful in real applications to predict its binding proteins if the RNA-specific prediction is formulated as a multi-label classification problem, whose inputs are RNA sequences and labels are available RBPs.
In this study, we present a multi-label deep learning method called iDeepM to predict a set of binding RBPs for RNA sequences. We first collect a set of RBPs and their RNA targets from CLIP-seq data. For each RNA, we consider its binding RBPs as its labels. Next, we train a joint model based on multi-label deep learning to predict how a RNA is attached by a set of RBPs. iDeepM uses a convolutional neural network to learn shared high-level features across RBPs, and long short term memory network (LSTM) under multi-label learning framework to infer the dependency and combinations of RBPs.
python 2.7
Keras 1.2 (https://github.com/keras-team/keras ) and its backend is theano 0.9
Sklearn v0.18 (https://github.com/scikit-learn/scikit-learn)
The data is downloaded from https://github.com/gianlucacorrado/RNAcommender/tree/master/examples, it contains 67 proteins and binding UTR sequences. we also require other UTR sequence to extract negative sequecnes, UTR_hg19.fasta is downloaded from http://aura.science.unitn.it/site_media/download/UTR_hg19.fasta .
Xiaoyong Pan^, Yong-Xian Fan, Jue Jia, Hong-Bin Shen^. Identifying RNA-binding proteins using multi-label deep learning. SCIENCE CHINA Information Sciences. 2019, 62(1).