Predicting The Association Between MiRNA And Disease Based On Matrix Completion

To date,38,589 miRNA molecules have been found in animals,plants and viruses.At the transcription and translation levels,mirnas inhibit protein production and gene expression by binding to the target 3 ’-UTRS.Mirnas affect a variety of biological processes,such as cell proliferation,differentiation,apoptosis,metabolism,aging,signal transduction and viral infection.Thus,disorders of miRNAs,including expression disorders,often play an important role in the occurrence and development of many diseases,including but not limited to cancer,cardiovascular diseases and neurodegenerative diseases.This further promotes the emergence of association prediction between miRNA and disease,which is a positive direction in the field of bioinformatics.At present,researchers have proposed many computational methods for predicting disease-related mirnas,which have many problems.The existing prediction methods can be divided into three categories :(1)methods based on score function,which depend on similarity calculation;(2)Methods based on complex networks or graphics algorithms,with high false positives and high complexity;(3)The method based on machine learning algorithm lacks negative sample training problem.Based on the analysis of existing methods,this paper studied the prediction methods of disease-related miRNA and proposed two disease-related miRNA prediction methods.The main work is as follows:(1)Prediction of potential disease-related miRNAs based on induction matrix completion method.Since the miRNAs functional similarity matrix contains zero value,this method USES the miRNAs functional similarity matrix to carry out the Gaussian interaction spectrum nuclear similarity to describe the relationship between miRNAs.Secondly,the main feature vectors of miRNAs and diseases were extracted by principal component analysis.The method then calculates the interaction profile of a new miRNA and its adjacent miRNAs.Finally,the method USES the main feature vectors and the constructed new miRNAs interaction spectrum,and USES the accelerated approximate gradient iteration method to complete the matrix,so as to obtain the incidence matrix.(2)Complete the model to predict disease-related Mi RNAs based on bayesian probability matrix.First,principal component analysis was used to extract the main feature vectors from the functional similarity matrix of miRNAs and the semantic similarity matrix between diseases.Then,bayesian probability method was used to obtain the respective factors,namely miRNA factor and disease factor,from the matrix composed of miRNAs feature vectors obtained in the previous step and the matrix composed of disease feature vectors.Finally,the correlation matrix between miRNA and disease was completed by using the learned miRNA factor and disease factor.