Passive absorption prediction of transdermal drug application with Artificial Neural Network

摘要

The modernization of IT leads to an improvement in numerous segments of the pharmaceutical activities. Thus, new field, pharmaceutical engineering was formed. Drug design is segment of it. Within that, the artificial neural network (ANN) possibility to determine the most suitable drug molecule for a transdermal administration is tested. Transdermal preparations industry is predicted to be worth cca 81.4 billion US dollars by 2024. This approach to drug design should be increasingly used due to convenience and cost-effectiveness of the method. The ANN was developed and trained by inserting over 500 dataset. The set included physiochemical parameters of drug molecule (molecular mass, melting point and partition coefficient), pharmacokinetic parameters (elimination half-time, dose and bioavailability) and biological parameters (skin permeability and coefficient of skin permeability). These parameters endeavored drug classification in one out of two success categories, respectively whether transdermal administration is possible or not. Various types of ANN were tested in order to acquire best accuracy and reliability. Types of ANN that resulted with accuracy above 95% have been considered. Sensitivity analysis of output variables related to input, suggested that certain parameters are more significant. Lastly, selected ANN had highest accuracy and least input parameters. Effect of overfitting was avoided while training selected ANN with highest accuracy. Based on the results of this study, ANN could be successfully used for predicting passive absorption of drug molecule through transdermal application. In addition, ANN could be used in order to facilitate processing large number of data. Hence, predicting mode of application for drag administration.