Deep Learning-based Target Detection For Drug-like Compounds

The discovery of the target of drug-like compounds has important role and significance in the development of new drugs.The target discovery technology of drug-like compounds based on bioinformatics and chemical informatics not only needs the comprehensive application of many related technologies and rely on the extensive storage and analysis of complex biochemical experiments and derived data,but also it requires professional background knowledge,and has a long development cycle and high cost.Therefore,there is a need for an efficient,rapid and costeffective method.Deep learning offers new possibilities for the discovery of drug-like compound targets.In this paper,deep learning is used to deal with the problem of drug-like compound target discovery by means of multi-label classification.A four-layer full-connection neural network is used as a deep learning model,in which the number of neurons in the input layer is 301,the number of neurons in both hidden layers is 1000,and the number of neurons in the output layer is 100.Because the distribution of target category labels in the original data set is extremely unbalanced,it belongs to the problem of unbalanced learning.In order to improve the degree of unbalanced distribution of target category labels in data sets and improve the classification performance of models,the multi label random oversampling algorithm was used to process data sets.Training in the CPU environment took 24 hours and its accuracy,precision,and recall were 70.5%,73.6%,and 75.3%,respectively.The results show that the use of deep learning to deal with the discovery of drug-like compound targets by multi-label classification is practical and feasible,and it can play an instructive role in the discovery of drug-like compound targets.However,due to the problem of too many target category labels in the data set and the extreme imbalance in the distribution of the target category labels present in the data sets,the number of target category labels that can be handled is relatively limited,and the classification performance needs to be improved.