Plant LncRNA-protein Interaction Prediction Using Deep Learning

Plant long non-coding RNA(lncRNA)plays a vital role in many biological processes mainly through interactions with RNA-binding protein.To understand the function of lncRNA,a basic method is to identify which type of protein interacts with lncRNA.However,although the conventional biological experiment method has high accuracy,it needs a lot of time and energy.If the experiment has no clear purpose,the efficiency is very low in the face of the huge amount of data.The proposal of machine learning model provides a good idea for solving such problems.It makes use of the computer's operation ability to predict first and then carry out biological experiments purposefully for verification,which greatly improves the work efficiency.However,the traditional machine learning model must select and extract features manually,which is greatly influenced by human factors,while the deep learning model can automatically extract features and classify them,further improving the accuracy of prediction results.In this paper,two deep learning models are proposed.The first model uses only sequence information,the stacked denoising autoencoder as the basic model,the gradient descent decision tree as the classifier,and uses logistic regression to fine-tune the results,named PLRPI.Experiments show that the results on Arabidopsis thaliana and Zea mays datasets(ATH948 and ZEA22133)are good,and the prediction accuracy is 90.4% and 82.6%,respectively.The second model uses sequence and structure information,and uses stacked denoising autoencoder and convolutional neural network as basic deep learning models and integrates the results,named PRPI-SC.The prediction accuracy was 88.9% and 82.6% on the same datasets.PLRPI and PRPI-SC also performed well on some public RPI datasets.These two models can accurately predict the interaction between plant lncRNAs and proteins,and each has its own focus,which plays a guiding role in the study of the function and expression of plant lncRNAs,and also has a strong generalization ability and good prediction effect on non-plant data.