Structure-based Prediction Of Drug Oral Absorption, Hepatic Clearance And Bioavailability

With the emergence of the genomics,proteomics and bioinformatics,novel therapy targets have been coming forth.In addition,with the aid of combinatorial chemistry and high throughput screening for the compound activity,the number of the new chemical entities and the candidate compounds are significantly increased.However,40%of the candidate compounds are rejected before the clinical test due to the poor pharmacokinetic properties.In order to evaluate the exploitability of the candidates,the pharmacokinetic properties should be known in the early stage of the drug discovery.And the prediction of pharmacokinetic properties should be considered in the early stage of the drug discovery, which is one of the main challenges in drug discovery.The structure based method,also as QSPR method,can be an efficient tool to predict important pharmacokinetic properties （such as drug absorption,clearance and bioavailability）.Based on this,the main purpose of our study is to explore the prediction of drug oral absorption,hepatic clearance,the unbound fraction in the microsomes and hepatocytes incubation,and the oral bioavailability with QSPR method.Furthermore,the identification of the important parameters in these models can be used to direct the drug design.The contents and results of our study are described as follows:1.Based on the membrane permeability and solubility related parameters,the QSPR model for F_a was established by the non-linear fitting method in Origin software.The QSPR model for F_a was constructed with logk_IAM,pH=7.0,MW,HD and logS_pH=7.0 values of 111 drugs.Among these drugs,nine drugs were considered as outliers due to the active transport and efflux phenomena.The results were described as follows:for the training set, n=70,R²=0.85,Q²=0.80;for the test set,n=24,R²=0.93.And 94%of the drugs could be correctly classified as low,medium and high absorbed drug（with the classification limits of 30%and 80%）.Furthermore,the logk_IAM,pH=7.0,MW and HD,representing the membrane permeability of drugs,and the logS_PH=7.0,representing the solubility of drugs were identified as important parameters in the F_a prediction model.2.The QSPR model for human CL_h prediction was built with 13 selected variables of 50 drugs.The model exhibited high performance,for the training set,n=36,R²=0.85,Q²=0.65, RMSE=0.28;for the test set,n=13,R²=0.73.One outlier was rejected due to its lowest CL_h value in the dataset.And 78%of drugs had the fold error values <2,just 10%of drugs with fold error>3 and with the average fold error（AFE） of 1.28.Furthermore,the LUMO, HOMO,Cosmic torsional energy and number of H-bond acceptors were identified as the important molecular descriptors to affect human CL_h.The QSPR model for CL_h values in rat was established with similar procedure to the model in human.The CL_h values of 10 aldose reductase inhibitors and 4 a-glucosidase inhibitors in rat were obtained from experiments.The predicted and the observed CL_h values in rat for these 14 compounds were compared.It was indicated that just 4 drugs had the fold error values >3（33%）.This external validation directly demonstrated that the performance of the CL_h model in rat was desirable.And the result indirectly indicated that the CL_h model in human was also exhibited high predictability.3.The QSPR model for fu_mic was successfully developed based on seven selected structural parameters of 86 drugs.The R² of the predicted and observed log((1-fu_mic)/fu_mic) for the training set（n=64） and test set（n=22） were 0.82（Q²=0.75） and 0.85 respectively.75%of drugs were found with fold error<2,only 2%of drugs with fold error>3 and AFE= 1.33. The predictive capability of fu_mic for neutral drugs compared well to that for basic compounds（R²=0.82,AFE=1.18） in our model.Our model appeared to perform better for neutral compounds when compared to models previously published in the literature.The lipophilicity,charge state,and the extent of ionization at pH 7.4 were identified as important properties affecting fu_mic.In addition,the QSPR model for fu_hep prediction was studied preliminarily.The R2 and Q2 for the entire dataset（n=37） were 0.62 and 0.51 respectively.81%of drugs were found with fold error<2,only 8%of drugs with fold error>3 and AFE= 1.41.The logD_7.0,the number of the total atoms,the number of the secondary amine groups,and the number of the negative charged groups were identified as important properties affecting fu_hep.Finally,the relationship between fu_mic and fu_hep was researched.It was indicated that there existed a strong linear relationship between fu_mic and fu_hep（n=37,R²=0.83）.Especially,the electrostatic interaction was important to the non-specific binding of drugs in microsomes and hepatocytes incubation.4.The absorption and hepatic clearance are two important parameters determining drug oral bioavailability.Based on our studies for the prediction of F_a and CL_h,the QSPR model for drug oral bioavailability prediction was built with 111 drugs.The stepwise regression and multiple linear regression methods were used to perform the variable selection and model construction.The predictability of the model was not desirable（for training set,n=82, R²=0.60,Q²=0.45,RMSE=21.06;for test set（two drugs were considered as outliers due to the pre-systemic metabolism and active transport）,n=29,R²=0.55）.While the model exhibited desirable predictive accuracy.79%of the drugs were found with fold error <2, only 13%of drugs with fold error>3,7%of drugs with fold error>5,3%of drugs with fold error>10 and AFE=1.64.About 70%of the drugs with medium or high F could be correctly classified.The false negative predictions were few,only 3%.There were same parameters among the F_a,CL_h and F models,such as logk_IAM,PH=7.0,LUMO and total dipole.However,the effects of these parameters on F_a/CL_h or F were different（positive or negative）.It was indicated that the prediction of F was complex.These results of our study could give new ideas for the further QSPR prediction of F.