Research On Drug Mining Method Based On Integrated Matrix Factorization Algorithm

With the rapid development of high-throughput genomics data,analyzing those data by applying effective data developing methods has been widely studied and used.At present,the main cure for cancer is drug therapy.Traditional drug discovery methods follow the“one drug-one target”line.But many complex diseases are caused by unique pathway functions instead of single genes or abnormal proteins.Thus,identifying drug-pathway associations is an important issue in drug discovery.In this paper,we find the disadvantages of traditional methods by summarizing and analyzing the results of research.The sparsity produced by L_2,1-norm penalty is too dispersive.Thus,based on the results of traditional research,that is the iPaD?Integrative Penalized Matrix Decomposition?method,this paper proposes three novel drug-pathway association identification methods:the L_2,1-Integrative Penalized Matrix Decomposition method(L_2,1-iPaD),the L₁L_2,1-Integrative Penalized Matrix Decomposition method(L₁L_2,1-iPaD)and the GL_2,1-Integrative Penalized Matrix Decomposition method(GL_2,1-iPaD).The penalty of L_2,1-norm can produce row sparsity.Based on the advantage of L_2,1-norm,the L_2,1-iPaD method is proposed to improve the performance of iPaD method by imposing L_2,1-norm constraint on the regularization term.The sum of L₁-norm penalty and L_2,1-norm penalty can produce inner row sparsity.Thus,we propose the L₁L_2,1-iPaD method to enhance the sparsity of drug-pathway association matrix by imposing L₁-norm and L_2,1-norm constraints on the regularization term.Manifold learning can keep the internal structure of data.The GL_2,1-iPaD method is proposed by adding manifold learning to identify drug-pathway association pairs.In order to verify the effectiveness of the three drug-pathway association identification algorithms,we apply those three algorithms to analyze integrated gene expression data and drug sensitivity data,and the existing algorithm as the control group.The experimental results show that these three algorithms are effective in drug-pathway association identification,and those algorithm performance is significantly higher than the traditional algorithms.The innovation points of this paper include three aspects:?1?this paper proposes a new drug-pathway association identification method(L_2,1-iPaD)to identify drug-pathway associations by adding L_2,1-norm penalty the regularization term;?2?based on L₁-norm and L_2,1-norm,the L₁L_2,1-iPaD method is proposed;?3?based on L_2,1-norm and manifold learning algorithm,the GL_2,1-iPaD method is proposed,and this method has been successfully applied to integrated gene expression and drug sensitivity data sets.