Predictive Activity Models Of Flavonoids

Objective The procedures of drug candidate discovery,pharmacologicical screening,pharmacodynamic and clinical research were a cost and time consuming procedure.In addition,with the development of combinatorial chemistry,millions of potential compounds were synthesized,while the traditional pharmacological research could not keep the space with the synthesis.The researches on absorption,distribution,metabolic and excretion of drug were the main work in the late stage of drug development,which also possesses that problem mentioned above.Quantitative structure activity relationship is a method that correlates the structural or property descriptors of compounds with the activities,therefore is applied for preliminary potential drug screening.Taking the multi-activity and diversity pharmacokinetics of flavonoids,into account,biopartitioning micellar chromatography,back propagation artificial neural network and counter propagation artificial neural network were used to construct pharmacokinetics property,affinity to P-glycoprotein and selective of estrogen receptor models of flavonoids,which could be helpful for the development of drug with structure of flavonoids.Content The pharmacokinetic quantitative retention activity relationship model,back propagation artificial neural network model of the affinity of flavonoids to P-glycoprotein and counter propagation artificial neural network model of the selective of flavonoid toward estrogen receptor were developed with the experimental activities reported in literatures.The performance of models were evaluated by statistics method.Methods(1)The pharmacokinetic quantitative retention activity relationship model of flavonoids:The retention time of the flavonoids were determined,and the retention data of flavonoids and their corresponding properties were adjusted to second-order polynomial QRAR model,then the predictive of model were evaluated by cross-validation,while the best condition of mobile phase and corresponding models were selected according to the statistic results.(2)The quantitative structure activity relationship models on the affinity to P-glycoprotein of flavonoids.Molecular descriptors calculated by PaDEL-Descriptor were subjected to variable reduction using the genetic algorithm and stepwise regression.Then selected descriptors were used for constructing linear and non-linear models,which were developed by back propagation artificial neural network and multiple linear regression respectively.Further more,the robustness and predictive of these models were compared.(3)The quantitative structure activity relationship models on the estrogen receptor selective of flavonoids.The estrogen receptor selective data were extracted from the literatures;the molecular descriptors were also calculated by PaDEL-Descriptor,which were reduced in number by Kohonen neural network.The selected descriptors and estrogen receptor selective data were then taken to construct counter propagation artificial neural network models of flavonoids.The structure of net model was optimization by genetic algorithm.The predictive of the model were evaluated by external validation set.Results(1)The pharmacokinetic quantitative retention activity relationship model of flavonoids.The goodness of fit of model developed in pH 7.4 is better than those developed under other pH conditions.In terms of dependent variable,the models of half-life and total clearance constructed by equation activity=a(logk)~2+b(logk)-c were better,while the model of apparent volume constructed with equation log(activity)=a(logk)~2+b(logk)+c was better in pH 7.4 Brij 35 condition.However,the models of half-life,total clearance and apparent volume constructed with equation activity=a(logk)~2+b(logk)+c were better in pH 7.4 Brij35/SDS(85:15)condition.(2)The quantitative structure activity relationship models on the affinity to P-glycoprotein of flavonoids.The correlation coefficient of the multiple linear regression model on affinity to P-glycoprotein about flavonoids was 0.855,while in analysis of variance,the Fisher ration shows a value 45.109.All the p-values for the descriptors lees than 0.05.The robustness,productiveness and applicability of the model were supported by the value of q~2(q~2=0.8138),internal predictive squared correlation coefficient(Rint2=0.7912)and predictive squared correlation coefficient(Rext2=0.6916).However,,with the genetic algorithm-partial least squares descriptor selection procedure in the non-linear model construction,the number of descriptors introduced to the artificial neural network model was final reached at 14.The model with the topology structure of 14-4-1 showed correlation coefficient of 0.9199,and extral predictive squared correlation coefficient of 0.8713.(3)The quantitative structure activity relationship models on the estrogen receptor selective of flavonoids.After the Kohonen descriptor selection procedure,the number of descriptors introduced to the artificial neural network model was final reached at 22.The counter propagation artificial neural networks model,which hold 12×12 structure and training epochs at 400 get the best prediction performance.The cross validation correct classification rate was 78%and 80%of sample was correct classification in external validation set.Conclusion The pharmacokinetic QRAR models of T1/2,Vd and Cl obtained in biopartitioning micellar chromatography showed good interpretative and predictive ability.The best condition for construct model was pH 7.4 0.05mol/L Brij 35/SDS(85:15)and pH 7.4 0.05mol/L Brij 35.The performance of 14-4-1 back propagation artificial neural network was better than multiple linear regression model,even better than the other 3D-QSAR model based on ligand and acceptor docking.The estrogen receptor selective model constructed by CPANN showed a good classification(80%)in flavonoids,which demonstrated CPANN model following descriptor calculayion with PaDEL-Descriptor as the calculation tool was promising in flavonoids estrogen receptors selective classification.