TIWMFLP: Two-Tier\nInteractive Weighted Matrix Factorization\nand Label Propagation Based on Similarity Matrix Fusion for Drug-Disease\nAssociation Prediction

Accurately identifying new therapeutic\nuses for drugs\nis crucial\nfor advancing pharmaceutical research and development. Matrix factorization\nis often used in association prediction due to its simplicity and\nhigh interpretability. However, existing matrix factorization models\ndo not enable real-time interaction between molecular feature matrices\nand similarity matrices, nor do they consider the geometric structure\nof the matrices. Additionally, efficiently integrating multisource\ndata remains a significant challenge. To address these issues, we\npropose a two-tier interactive weighted matrix factorization and label\npropagation model based on similarity matrix fusion (TIWMFLP) to assist\nin personalized treatment. First, we calculate the Gaussian and Laplace\nkernel similarities for drugs and diseases using known drug-disease\nassociations. We then introduce a new multisource similarity fusion\nmethod, called similarity matrix fusion (SMF), to integrate these\ndrug/disease similarities. SMF not only considers the different contributions\nrepresented by each neighbor but also incorporates drug-disease association\ninformation to enhance the contextual topological relationships and\npotential features of each drug/disease node in the network. Second,\nwe innovatively developed a two-tier interactive weighted matrix factorization\n(TIWMF) method to process three biological networks. This method realizes\nfor the first time the real-time interaction between the drug/disease\nfeature matrix and its similarity matrix, allowing for a better capture\nof the complex relationships between drugs and diseases. Additionally,\nthe weighted matrix of the drug/disease similarity matrix is introduced\nto preserve the underlying structure of the similarity matrix. Finally,\nthe label propagation algorithm makes predictions based on the three\nupdated biological networks. Experimental outcomes reveal that TIWMFLP\nconsistently surpasses state-of-the-art models on four drug-disease\ndata sets, two small molecule-miRNA data sets, and one miRNA-disease\ndata set.