An Lncrna-disease Association Prediction Method Based on Nearest Neighbor Matrix Completion Model

Abstract

Abstract To address the problem that the traditional correlation metric defaults the 0 terms in the adjacency matrix of lncRNA-disease to irrelevant, ignoring the fact that it is only not validated now but maybe validated as relevant in the future, we propose a correlation calculation method that incorporates potential correlation properties, which is verified by comparative experiments to have better performance than the traditional method in the model; meanwhile, with the idea of nearest neighbor, we design the matrix completion model (DMWNN) to reassign values to the 0-terms in the adjacency matrix. The correlation is used instead of the traditional Euclidean distance to screen more valuable neighbors; combined with the linear decay strategy of distance weights, the interference of low correlation data is reduced, and thus the accuracy of reassignment is improved. The AUC value reached 0.9480 in the five-fold cross-validation experiment and 0.9603 in the leave-one-out cross-validation experiment, and the experimental results showed that the DMWNN model can effectively explore the potential association between lncRNA and disease and has strong predictive performance.