Prediction Of Drug Properties Based On Deep Learning Framework

There are currently many problems such high forecasting costs and poor prediction results in the prediction of drug discovery.So how to solve such problems has become one of the hot points for scientific researchers.Predicting drug penetration is the key to assessing the ability of drugs to be well absorbed.Artificial determination involves a large number of manual interventions,and the data sets used in the predicted data model are small.These factors are prone to cause the complex feature selection and the disadvantage of overfitting.In the prediction of drug toxicity,although using the random forest(RF)and other machine learning models,high computational costs and other issues.Neural network is the traditional method to deal with classification and regression.In recent years,through the continuous improvement and updating,it gradually overcome the limitations of the algorithm itself to improve the efficiency of the algorithm.Among them,the deep neural network shows a strong ability of autonomous learning and has achieved good predictions in some application areas.Based on the above background,this paper used the depth neural network learning method to research the problems of drug permeability and toxicity prediction by the construction of the depth of the neural network framework and the use of large data sets.The main work includes:(1)In view of the higher ability of the neural network,the UG-RNN molecular coding method is used to study the optimal UG-RNN network structure.(2)A binary model based on the depth neural network is proposed to predict the drug(LDA)and gradient lifting tree(GBT)method.The result shows that the binary classification models are superior to the contrast models in the prediction accuracy.(3)The two-Based on the depth learning structure,the regression model of drug toxicity was developed.Multiple sets of data sets were constructed in the regression experiment,and the final experimental results were compared with the single-layer neural network(NN).The result shows that the learning performance and the prediction accuracy of the regression performance model are better than the NN's;(4)Combining the depth neural network framework with the support vector regression,we got the combined regression model and tested its performance.By evaluating the prediction accuracy,it shows that the learning method of the set depth has more powerful prediction ability.In the experimental part,a large amount of data is collected in this paper,and eight sets of data sets are processed to detect the model performance.Among them,663 compounds and 209 molecular characteristics were found in the drug absorption performance.In the drug toxicity prediction experiment,475 groups of drug data were used as training sets and 198 groups of drug data as test sets.In order to improve the performance of the algorithm,the drop-out method is used to reduce the overfitting problem,and the modified linear unit(ReLU)method is used to reduce the gradient disappearance problem.Through the discussion and comparison of the prediction results and the performance of the classification model and the regression model,it is shown that the prediction results of the depth learning are superior to the other machine learning prediction models.