Study On Predicting Absorption Properties And Human Pharmacokinetic Profile For Oral Drugs

Aims To construct models to predict important absorption properties and human PK parameters to biopharmaceutical analysis and drug development for oral drugs.To predict human concentration-time profiles finally based on simple PK model and predicted pharmacokinetic parameters before conducting clinical experiments.Methods1.A model to predict Caco-2 permeability based on support vector machine,CDK molecular descriptors,and a Caco-2 permeability data set of 100 drugs(training set: 77 drugs;test set:46 drugs)was constructed and validated.Caco-2 permeability experiment for Danshensu was conducted to validate the built model.2.A model to identify P-gp substrates based on random forest,CDK molecular descriptors,and a data set of 212 compounds(training set:170 compounds;test set: 42 compounds)was constructed and validated.3.A model to classify drugs into oral absorption classes based on random forest, Dragon(?)molecular descriptors,and a data set of 234 drugs(training set:188 drugs; test set:46 drugs)was constructed and validated.4.A model to predict human apparent volume of distribution based on genetic algorithm, MLR,CDK molecular descriptors,and a 122-drug data set(rat,dog,monkey,and human apparent volume of distribution data)was constructed and validated.5.A model to predict human clearance based on genetic algorithm,support vector machine,CDK molecular descriptors,and a 122-drug data set(rat,dog,monkey,and human clearance data)was constructed and validated6.Human pharmacokinetic data for 10 drugs was collected.Pharmacokinetic parameters for these 10 drugs were predicted.Pharmacokinetic curves were simulated based on these parameters and one compartment oral absorption model.Results1.The correlation coefficients of the experimental and predicted Caco-2 apparent permeability for the training set and the test set were 0.88 and 0.85,respectively. Membrane permeability of compounds was determined by number of H-bond donors and molecular surface area properties.The predicted Caco-2 permeability for Danshensu kept accordance with the experimental values.2.An external test data set which contained 42 compounds(24 P-gp substrates)was employed.The misclassification rates for P-gp substrates,non-substrates and the total compounds on the test set were 12.50%,16.67%and 14.29%,respectively. Leave-one-out cross-validation on the total compounds was 22.6%.The study on the relation between CDK descriptors and P-gp substrates presented that sum of the atomic polarizabilities and charged partial surface area played an important role in identifying P-gp substrates.3.The absorption classification accuracies of the RF model on the training set(188 drugs)and the test set(46 drugs)were 98.89%and 86.96%,respectively.The importance of molecular properties quantified by the RF model such as PSA,logP,etc. kept in accordance with the viewpoints of previous studies.The RF model was also validated to be insensitive to active transport drugs.4.The values of cross-validated correlation coefficient and relative root mean squared error(rRMSE)for leave-one-out cross validation of the final human apparent volume of distribution MLR model were 0.6495 and 0.3968,respectively.Genetic algorithm selected 10 important molecular descriptors to allometric scaling up problem. "Drug-like" properties played an important role in predicting human apparent volume of distribution.5.The values of cross-validated correlation coefficient and relative root mean squared error(rRMSE)for leave one out cross-validation of the final human clearance SVM model were 0.6120 and 0.4061,respectively.Genetic algorithm selected 10 important molecular descriptors to allometric scaling up problem."Drug-like" properties did not play an important role in predicting human clearance.6.Human pharmacokinetic simulation for 10 drugs was implemented.In these 10 simulations,8 were acceptable while 2 were insufficient.Conclusion The absorption property models(Caco-2 permeability,P-gp substrate)and the PK parameters models(human oral absorption,human apparent volume of distribution,and human clearance)achieved good performances.These models can serve drug development independently and efficiently.The idea to predict human concentration-time profiles based on animal data and chemical structure was implemented finally.The methods and conclusions in this thesis are valuable to biopharmaceutical analysis research,human PK studies,and drug development.