Prediction Of Plant NcRNA Interactions Based On Multi-scale CNN And BiLSTM

Mutual regulatory mechanisms between non-coding RNAs(ncRNA)play important roles in many animal and plant life activities,such as cell growth,differentiation,and proliferation.At present,the research on the interaction between miRNA and long non-coding RNAs(lncRNA)has been relatively more in animal and human diseases.It provides new solutions for the diagnosis and treatment of animal and human diseases,and there is relatively little research in plants.Therefore,the study of plant miRNA-lncRNA interactions can not only analyze the biological functions between genes,but also provide new ideas for plant genetics breeding.Traditional methods have biological identification and feature engineering,but biological identification has a high cost and much time,and cannot be identified in large quantities;feature engineering involves too much manual intervention and complicated extraction processes.Therefore,using deep learning methods to predict the interaction of ncRNA in plants has great advantages.This paper presents a multi-scale convolutional neural network(CNN)and bidirectional long short-term memory neural network(BiLSTM)model of plant ncRNA interaction prediction model(MCMPLA)that introduces the attention mechanism.It uses multi-scale CNN and multi-pooling operations to enrich feature diversity,and BiLSTM is used to solve the problem of long-term missing information dependence.At the same time,the attention mechanism is introduced to obtain key features to achieve the purpose of efficiently and accurately predicting the interaction relationship.First,the clustering undersampling operation is performed on the negative set to ensure that the positive and negative samples are balanced.To preserve the dependent relationship between adjacent nucleotides,the sequence is encoded by k-mers method.Then,CNN of multi-scale convolution kernel and multi-pool are used instead of single-scale cases to extract topic features of different lengths,thereby enriching feature diversity.At the same time,BiLSTM is used to solve the problem of long-term dependence of sequence information.Finally,an attention mechanism is introduced.Different features are assigned weights to obtain key features and further optimize the model.In order to verify the advantages of the proposed model MCMPLA,based on zea mays,solanum tuberosum and triticum aestivum datasets,experiments respectively compared feature-based machine learning methods,deep learning models of different architectures,andexisting methods,showing that MCMPLA is superior to existing methods and traditional machine learning methods,and validating the validity and rationality of MCMPLA.