MiRNA And Disease Association Prediction Algorithm

MicroRNAs(miRNAs)are single-stranded small RNAs of about 21 to 23 bases in size,and are produced by the enzyme Dicer from a single-stranded RNA precursor with a hairpin structure of approximately 70-90 bases in size.In recent years,more and more studies have shown that miRNAs play an important role in various biological processes,and the expression of miRNAs themselves and the regulation of genes may affect various diseases.Therefore,it is very important to identify the relationship between miRNAs and diseases,and it has become a research hotspot in recent years.The early research methods were mainly based on biological experiments.The accuracy of the obtained results was high,and the relationship between miRNAs and diseases actually existed was fundamentally verified.However,the experimental methods have the disadvantages of high cost,long time-consuming,low success rate,and researchers are committed to finding more efficient calculation methods to solve the problem.With the increase of known associations,the use of known associations to use computational biology to predict the association between miRNAs and disease seems to have become a breakthrough for researchers.On the one hand,its results can eliminate a large number of “false answers” and save valuable experimental costs.On the other hand,good biometrics can even replace biological experiments,and use a very high accuracy rate to predict the association between miRNAs and diseases..This paper divides the current calculation methods for miRNAs-disease associations into two categories:(1)a method based on network topology,and(2)a method based on machine learning.The former further predicts possible "edges" by establishing associations between miRNAs and nodes in the disease network,ie,new mi RNA-disease associations,but the experimental results of this method depend on a high-confidence biological network model.,and can not be effective on new miRNAs or new diseases;the latter applies machine-learning related algorithms to solve this problem,and solve the problem of prediction of new miRNAs and new diseases,but at the same time,such methods need to solve feature extraction and negative cases.There are two major issues missing.In addition,in these two types of calculation methods,it was found that the model used the MISIM database to calculate the similarity of miRNAs,whereas the MISIM database was derived from the association of miRNAs and diseases,which resulted in cross-validation.Some unnecessary logic paradoxes.Based on the problems in the above methods,this paper proposes a sequencebased similarity modeling method for miRNAs and a computation model based on PU-learning-based computational models AIWC(Adjustable Iterative Weighted Classifier)and LMFMDA(Least Squares Optimization Matrix Factorization method for mi RNA-Disease Association).AIWC is based on the assumption that the set of tagged positive tags is randomly selected from the true population set.The PUlearning correlation algorithm is used to model the association prediction of miRNAs and diseases,focusing on the recall of known samples and optimizing the entire model.The correlation forecast results.LMFMDA applies a matrix factorization algorithm to solve this problem.It projects miRNAs and diseases into the latent variable space,fits the existing miRNAs-the disease association matrix,and constrains the miRNAs similarity matrix to the disease similarity matrix to obtain miRNAs and The disease is represented in the latent variable space to predict new miRNAs-disease associations.Unlike conventional practice,we introduce complementary miRNAs and disease variables to ensure convergence to optimal solutions during optimization.This method has a good effect on new miRNAs and diseases,and its AUC value can reach 0.8511,showing obvious superiority to the existing high-efficiency methods.