McBel-Plnc: A Deep Learning Model for Multiclass Multilabel Classification of Protein-lncRNA Interactions

Abstract

One main function of long non-coding RNAs (lncRNAs) is to act as a scaffold facilitating multiple proteins to form complexes. Most of available prediction models for protein-RNA interactions, however, were proposed as a binary classifier, which limited on predicting the interaction between the non-coding RNAs and each individual RNA-binding protein (RBP). Hence, to predict if a lncRNA is acting as a scaffold, we consider this problem as a multiclass multilabel classification problem. To solve this problem, the high confident CLIP-seq data were selected from the POSTAR2 database with an augmentation of the data for the RBP classes with a small number of interacting lncRNAs. We then constructed a deep learning model for multiclass multilabel classification, called McBel-Plnc, based on the convolutional neural network (CNN) and long-short term memory (LSTM) using each of the five datasets randomly generated from the prepared data. Based on macro average, the test results showed the high precision of 0.9151 ± 0.0038 averaged from the five models with the lower recall of 0.5786 ± 0.0208. The small standard deviations confirmed the model stability. Comparing with iDeepE with a binary relevance method, iDeepE got the higher recall with the significantly lower precision (0.6912 and 0.1987, respectively). This result suggested that our model is competent to predict the protein-lncRNA interactions, especially with the lncRNAs targeted by multiple proteins. This suggested the potential to infer the insights of lncRNA functions and molecular mechanisms.