Identification of Human LncRNA-Disease Association by Fast Kernel Learning-Based Kronecker Regularized Least Squares

Abstract

As the function of lncRNA is gradually understood, they have been found regulating the expression of target genes at the post-transcriptional level, and their abnormal functions may lead to so many diseases. Then, identifying the lncRNA-disease associations (LDA) can help to better understand its pathogenesis, promote the search for biomarkers of disease diagnosis, and effectively prevent disease. To break through the limitations of the existing computational models, we put forward a novel computational method of lncRNA-disease association identification by employing Fast Kernel Learning with Kronecker Regularized Least Squares (FKL-KronRLS-LDA). This model first extracts three different similarity kernels in disease and lncRNA space respectively. Next, it fuses these distinct kernels into an integrated kernel with the optimized combining weightings indicating their importance. It then combines lncRNA kernel and disease kernel into one larger kernel by Kronecker product kernel. Finally, it adopts the regularization least squares to identify potential associations. In experiments of Leave one out cross validation (LOOCV) and 5-fold cross validation (5-fold CV), FKL-KronRLS-LDA respectively obtains an AUC of 0.917 and 0.856, which outperform other excellent computational models. Furthermore, in the case studies, 9, 8 and 8 out of top 10 identified lncRNAs are successfully confirmed by recent published literature for lung cancer, breast cancer and gastric cancer, respectively. In a word, FKL-KronRLS-LDA can effectively identify potential lncRNA-disease associations for human beings.