Prediction of Plant IncRNA-Protein Interactions Using Sequence Information Based on Deep Learning

摘要

Plant long non-coding RNA (IncRNA) plays an important role in many biological processes, mainly through its interaction with RNA binding protein (RBP). To understand the function of IncRNA, a basic step is to determine which proteins are interacted with IncRNA. Therefore, RBP can be predicted by computational approaches. However, the main challenge is that it is difficult to find interaction patterns or primitives. In this study, we propose a method based on sequences to predict plant IncRNA-protein interaction, namely PLRP1 uses k-mer frequency feature for RNA and protein, stacked denoising autoencoder and gradient boosting decision tree to learn the hidden interaction between plant IncRNAs and proteins sequences. The experimental results show that PLRP1 achieves good performance on the test datasets ATH948 and ZEA22133 based on IncRNA-protein interaction of Arabidopsis thaliana and Zea mays. Our method gets an accuracy of 90.4% on ATH948 and 82.6% on ZEA22133. PLRP1 is also superior to other methods in some public RNA-protein interaction datasets. The result shows PLRPI has strong generalization ability and high robustness. It is an effective model for predicting plant IncRNA-protein interactions.