Dsnpcmf: Predicting mirna-disease associations with collaborative matrix factorization based on double sparse and nearest profile

Lately, on account of being associated with many human diseases, more and more attentions are paid to microRNAs (miRNAs). Accumulating experimental studies of predicting novel miRNA-disease associations (MDAs) are costly and time-consuming. And there are also many unknown associations between miRNAs and diseases. Therefore, it is a momentous topic to predict possible associations between miRNAs and diseases. Also, it is urgent to increase the accuracy of predictive performance. In this paper, we put forward a computation method of Predicting MiRNA-Disease Associations with Collaborative Matrix Factorization based on Double Sparse and Nearest Profile (DSNPCMF) to estimate underlying miRNA-disease associations. In this model, we integrate Nearest Profile (NP) and Gaussian Interaction Profile (GIP) kernels of miRNAs and diseases to augment information of their neighbors and kernel similarities to improve the predictive ability. In addition, L_2,1-norm and L₁-norm are introduced into this method to increase the sparseness. Then five-fold cross validation is used for assessing our developed method. At the same time, simulation experiment is used to detect the result of prediction, including both known MDAs and new MDAs that are in descending order. In the end, the results prove that the accuracy of our prediction is better than other previous perfect methods. And our method has the ability to predict latent associations of miRNAs and diseases.