Prediction Of Human Oral Bioavailability And Half-life Method With Machine Learning Methods

The quality of AMDET is a key factor in determining the fate of a new drug in clinical trials.Traditionally,ADMET properties of drug candidate small molecules have been measured using expensive and time-consuming experimental tests,and using computational models to assess drug properties prior to new drug synthesis would benefit the drug development process,saving time and money.1.We collected a total of 1588 drug molecules with human oral bioavailability data from the literature for the development of a consensus predictive classification model consisting of five random forest models.In this study,F=20% and F=50% were used as classification cutoffs for the classification of the data,and the consensus model showed excellent prediction accuracy on the independent test set of the two classifications.At the same time,the importance analysis of the input variables allows us to fully realize that some major molecular descriptors can significantly affect the HOB classification results of the model.2.We use a training set containing 1064 half-life data,using 12 2D,3D descriptors in a larger chemical space,inspired by the first work,using random forest,GBDT,XGBoost to vote with the same weight to build a consensus model for qualitatively predicting the half-life of small molecules of candidate drugs.Through the analysis of the prediction results of the model in a single sample and the whole sample,we can better understand the reasons that affect the half-life of small molecules.All of the above work is available as a web server for rapid assessment of human oral bioavailability or half-life of small molecules at www.icdrug.com/ICDrug/ADMET.The results of this study provide an accurate and easy-to-use tool for HOB or half-life based drug candidate screening,which can be used to reduce experimental costs and the risk of late-stage failure in drug development.