Prediction Of Disease Associations Network Based On Multiple Kernel Learning

For a long time,diseases have seriously affected human normal life activities and endangered people’s life and health.Therefore,it is very necessary to study the disease.Understanding the genes related to diseases is of great help in the prevention and treat-ment of diseases.However,it takes a long time and cost to determine the relationship between diseases and genes through biological experiments.In addition,the treatment of diseases depends on specific drugs.Compared with the development of new drugs,drug relocation has the advantages of using low-risk compounds and reducing the over-all development cost.Machine learning model has been widely used to explore the genetic information related to complex diseases.Similarly,inferring the association in drug disease network through the model provides effective help for drug reposition-ing.Therefore,it has important research significance and practical application value to predict disease network association by using computational model.In the research of gene-disease network link prediction,this paper proposes a Dual Hypergraph Regularized Least Squares model to predict the association of genes and diseases.The two feature spaces are described by using multiple similarity matrices of genes and diseases.At the same time,the central kernel alignment method is used to calculate the weight of the kernel matrix,and multiple kernel matrices are fused by lin-ear combination.In addition,hypergraph is also applied to graph regular terms to obtain high-order information of genes and diseases to improve prediction performance.In or-der to prove the effectiveness of the method proposed in this study,the model is tested on six benchmark data sets and genetic disease association network.Compared with the existing methods,the experiments show that the model proposed in this study has good prediction performance.In the research of drug-disease network link prediction,this paper proposes a Multiple Kernels Dual Graph Regularized Least Squares model to identify potential drug-disease associations.Different from the traditional Multi-ple Kernel Learning,the model integrates the multiple kernel learning process into the Laplace least squares framework.Firstly,multiple kernels of drug space and disease space are calculated respectively.Then,the model is constructed by using multiple ker-nel matrices and related Laplace regularization terms.Finally,the weight parameters of multi-kernels and Laplace least squares are optimized by alternating least squares algorithm.The model is superior to the traditional multi-kernel learning method and existing models on three real drug-disease association datasets.In addition,this paper also uses a case study to test the ability of the model to predict new associations.In this paper,two models based on Multiple Kernel Learning are proposed to pre-dict disease-related networks respectively.The experimental results show that the model has excellent performance in relevant datasets.The research of this paper shows that using hypergraph Laplace regularization to obtain the high-order geometric structure information between data can improve the prediction effect of the model;If it is feasi-ble to add the multi-kernel learning process to the loss function for joint optimization;Regular graphs and L₂regularization constrains variables in a complementary relation-ship.