Computer-aided Prediciton of ADME Properties of Drugs
|Keywords||Pharmacokinetic properties Quantitative structure property / activity relationships Genetic algorithms - Multiple linear regression Least squares support vector machine method|
Since the late 1990s, the drug development studies have shown that one very important reason leading to the failure of expensive drugs developed in the late adverse drug metabolism kinetics and animal toxicity (ADMET) nature. Pharmacokinetic properties reflected the absorption of the drug in vivo (absorption, A), distribution (distribution, D), metabolism (metabolism, M), and excretion (excretion, E) the law. Drugs early ADMET properties of human tissue functional protein drug targets, combined with in vitro studies with computer simulation method, to explore the role of drugs in vivo. The drugs early ADMET properties evaluation method can significantly improve the success rate of drug development, reduce the cost of drug development, reduce the incidence of drug toxicity and side effects, and can guide rational drug. Thus, theoretical prediction the ADMET nature is of great significance. The various metabolic kinetic properties of the drugs is determined by its chemical structure and physicochemical properties. Change of chemical structures and physicochemical properties of the drug can lead to changes in the biological activity. Quantitative structure-activity relationship (quantitativestructure-activity/property relationship, QSA / QSPR) is some mathematical statistical methods to establish a quantitative relationship between the physiological activity of the compound or some kind of the nature of their molecular structure. Through these quantitative relationship, we can predict the physiological activity of the compound, or some properties, guiding the synthetic design a compound having a higher activity. QSAR is the most widely used drug design one of the methods. The first chapter of this thesis, the principle of drug ADMET properties of introduction and quantitative structure activity / property relationship (QSAR / QSPR) method. The following brief we use QSPR / QSAR the theoretical methods ADMET three: Chapter II QSPR method studies the relationship between the 60 drug-like compounds inherent water solubility and its structure. GA selected the five parameters, and establish a genetic algorithm - multiple linear regression (GA-MLR) and least squares support vector machine (LS-SVM) model with five parameters. The resulting model through a variety of authentication methods, including leave one out (LOO) cross-validation, Y-random verification as well as external test set validation. GA-MLR model RMSR training set was 0.51, and the coefficient of determination, R ~ 2 0.88, the cross-validation coefficient O ~ 2 0.84, and Y-random R ~ 2 0.07; results of the test set R ~ 2 0.82, RMSE 0.84. LS-SVM model training set O to 2 was 0.86, of R ~~ 2 0.90, RMSE 0.47; tests set R ~ 2 0.83, RMSE 0.49. Visible GA-MLR and LS-SVM method results close to these two models are robust and can be used to accurately predict the inherent water solubility of the compound. GA-MLR forecast Chapter 70 is closely related to the central nervous system (CNS) drug microemulsion electrokinetic chromatography (MEEKC) capacity factor and the permeability of the blood-brain and the molecular structure and capacity factor made a correlation analysis, and thus indirectly to predict drug brain tissue non-binding rate for the prediction of blood-brain barrier to provide a simplified and easy research ideas. The model test set, leave one out cross validation, Y-randomization method validation. The evaluation of the training set of the model were: RMSE 0.25 coefficient of determination (R ~ 2) 0.87, cross-validation (Q ~ 2) 0.75 and Y-random R ~~ 0.10: test set decision coefficient R ~ 2 0.91, RMSE 0.23. As can be seen, the GA-MLR model has good predictive ability. The fourth chapter discusses the theoretical prediction of 223 drug-like compounds of olive oil solvation free energy. GA-MLR and LS-SVM model with the GA elected five variables were established. GA-MLR Y-random R ~ 2 was 0.03. GA-MLR model and LS-SVM model training set O to R ~~ RMSE, AARD, F values ??are very close to 0.88 vs 0.87,0.89 vs 0.89,0.67 vs 0.67,20.08% vs 19.80%, respectively, 271.86vs 275.60; model of GA-MLR and LS-SVM model test set R to 2, RMSE, AARD, F are basically similar, ie 0.91 vs 0.92,0.73 vs 0.74,15.26% vs 14.62%, 82.86 vs. 84.92: means that the GA-MLR model and LS-SVM model predictive ability is relatively similar, using a linear model to predict the olive oil solvation free energy. Chapter 1080 organic compounds diamagnetic susceptibility quantitative structure property relationship (QSPR). GA elect three most important descriptor to create multiple linear regression (GA-MLR) model, and then leave one out cross validation (LOO) Y-random testing, as well as external test set to verify the model has good stability. GA-MLR model training set of evaluation were: AARD 4.51%, RMSE was 5.63, and the coefficient of determination (R ~ 2) 0.98, cross-validation (Q ~ 2) 0.98 and Y-random R to -0.02; evaluation of the test set are as follows: the coefficient of determination, R ~ 2 0.98, RMSE 7.14, and the AARD 5.24%. In comparison, the source literature training set R ~ 2 0.98, RMSE 5.41, and AARD 5.07%, while the test set R ~ 2 0.98, RMSE 6.05 and AARD 5.52%. GA-MLR model gives the the AARD value smaller, so the GA-MLR model prediction accuracy of the source literature is similar.