Disease-related MiRNA And Anticancer Peptides Prediction Research Based On Neural Network

Complex disease such as cancer is the main killer threatening human health,but the complete cure of such disease is currently a research problem that human cannot conquer.In recent years,a variety of small non-coding RNAs(mi RNAs)have been considered to be closely related to the generation,deterioration,metastasis and other processes of complex diseases.On the other hand,related studies on anticancer peptides have shown that this kind of polypeptide is expected to inhibit or even cure cancer.Based on the computer simulation,the disease-related mi RNA prediction model and the anticancer peptide prediction model can help identify mi RNAs and anticancer peptides with the most potential for efficacy,which is expected to reduce large-scale biological experiments,thus saving resources and speed up the development of treatments for complex diseases.In this paper,two models based on neural network were constructed to predict complex disease-related mi RNAs and anticancer peptides respectively.Firstly,this paper proposes a model called WDLMDA(Wide and Deep Learning for mi RNA-Disease Association Prediction),which combines neural network and linear regression method,to predict disease-related mi RNAs.In this model,mi RNA similarity and disease similarity were used to construct feature data for disease-mi RNA association pairs.Based on these feature data,a neural network model and a linear model were trained respectively.Finally,the prediction results of the association between disease and mi RNA were obtained by weighting the scores of the two models.Linear model is good at memorizing historical data,while neural network model has excellent generalization ability.Combining the advantages of the two models,the accuracy of WDLMDA in leave-one-cross verification and 5-fold cross verification exceeds the experimental results of most predecessors.In the specific case analysis,WDLMDA predicted the top50 related mi RNAs for esophageal tumor,breast tumor and hepatocellular carcinoma,and most of them have been verified by relevant literatures or databases.In addition,a neural network model called ACP-SA(Anti-cancer Peptides prediction based on Seq2 Seq model with the Attention mechanism)is proposed to predict anticancer Peptides.The model first used word2 vec algorithm to obtain the word vector representation of the sample sequence data and then calculated the feature representation of the physicochemical properties of amino acids and polypeptides.Based on these constructed feature data,a encoding and decoding model composed of bidirectional recurrent neural network and attention mechanism was trained,and finally the classification results of anticancer peptides were finally obtained through the output of the model.This model intelligently links the problem of peptide sequence classification with the problem of text classification,and achieves better experimental results than the previous work in the 5-fold verification on different datasets.