Prediction Of Drug-Target Interactions Via Feature Selection

With the rise of Artificial Intelligence,the development of drugs using machine learning methods can greatly reduce costs and time.Drug development is a complex and lengthy process,usually requiring years or even longer for research and development.It also requires a large investment and resource support,as well as professional knowledge,technology,personnel and other elements.Predicting drug-target interactions is an important part of the drug development process.Therefore,researchers urgently need accurate and fast prediction of drug-target interactions to develop effective drugs in innovative ways to overcome these shortcomings.This paper establishes two matrix decomposition models based on feature selection to predict drugtarget interactions,and the main research content is:Firstly,a prediction method based on multi-view neighborhood regularized logical matrix factorization.By utilizing the multi-view information in the drug and target space,the corresponding feature similarity matrix is constructed,and the corresponding Laplacian matrix is obtained according to the obtained similarity matrix.The obtained Laplacian matrices are fused as the condition of matrix decomposition,which improves the disadvantage of single feature possibly leading to incomplete information.From the experimental results,it can be seen that the multi-view fusion method proposed in this paper has good performance in predicting drug-target interactions.Secondly,a method of neighborhood regularized logical matrix factorization using neural tangent kernel to extract features was proposed.This takes advantage of the deep learning model’s ability to automatically extract features,improving the potential key information loss caused by traditional machine learning manually selecting features.The neural tangent kernel was used to extract the feature matrix of drugs and targets to optimize the prediction of drugtarget interactions.From the experimental results,it can be seen that the features extracted by the proposed neural tangent kernel model are significantly better than those selected manually,and have higher competitiveness.In addition,this paper further verifies the effectiveness of the method by analyzing the model’s ability to predict new drug-target interactions.