Theoretical Studies On The Prediction Of Drug-likeness And Oral Bioavailability

In order to obtain higher quality drug leads from a huge pool of compound library and improve the precision and efficiency of drug design, medicinal and computational chemists proposed the concept of drug likeness. A lot of rules and theoretical models have been developed to predict drug likeness based on molecular properties and structures. Meanwhile, it is well-known that oral bioavailability is a good indicator of the capability of the delivery of a given compound to the systemic circulation by oral administration, and an essential parameter in drug screening cascades as well as an important parameter related to drug likeness. Therefore, the reliable prediction of drug likeness and oral bioavailability play an important role in drug design. The goal of the thesis is to develop the reliable prediction models of drug likeness and oral bioavailability for promoting the efficiency of drug discovery.The thesis is divided into two parts. In the first part, we constructed the theoretical models for drug-likeness based on molecular properties and fingerprints by using Bayesian classification technique. First, we systematically examined the influence of different molecular properties and fingerprints on the prediction accuracy of drug-likeness. We found that the prediction accuracy of the Bayesian classification model based on simple molecular properties was not satisfactory and but the introductions of molecular fingerprints could improve the prediction accuracy obviously. Secondly, we constructed a variety of Bayesian classifiers by changing the ratio of drug and non-drugs in the training set and the size of the training set. Our calculation results show that the prediction accuracy of the Bayesian classifiers improved by increasing the size of the training set and the Bayesian classifiers based on the balanced training sets were better than those based on the unbalanced training sets. Finally, the recursive partitioning (RP) models were constructed and compared with the Bayesian classifiers, and we found that the Bayesian classifiers performed better than the RP models.In the second part, we collected a human bioavailability database with 1014 drug and drug-like molecules. Based on the extended dataset, systematic examinations of the relationships between various physicochemical properties and oral bioavailability were carried out to investigate the influence of these properties on oral bioavailability. Moreover, the genetic function approximation (GFA) technique was employed to construct the multiple linear regression models of oral bioavailability by using molecular fingerprints as the basic parameters, together with several important molecular properties. The results show that the simple rules based on molecular properties cannot give accurate predictions for oral bioavailability; however, the prediction models based on important molecular properties and molecular fingerprints given by GFA can give reasonable predictions. For the training set, the best model is able to predict human oral bioavailability with an r of 0.79, a q of 0.72 and a RMSE of 22.30%, and for a separate test set of 80 compounds, the model can predict the bioavailability with rtest=0.71 and RMSE = 23.55%. The studies also show that some important molecular fingerprints are closely related to important intestinal absorption and well-known metabolic processes.