Prediction of the penetration of drugs by Artificial Neural Networks

In this study an Artificial Neural Network was developed to predict the penetration of drugs through a polydimethylsiloxane membrane by molecular descriptors. A total of 245 drugs and their absorption experimentally determined values were arranged into various data sets to perform training and validation of the different implemented neural networks. Logarithms of the maximum steady-state flux (log J) values were correlated with four input variables; i) Count fo H-Acceptor Sites (CHA), ii) H-Donors Charged Surface Area (HDCA), iii) Gravitational index (G_b) and iv) Weighted Positive Charged Partial Surface Area (WPSA-2). Besides these, other neural networks were implemented with an extra input variable, which measuring the similarity of the different drugs (Distance). All developed neural networks present a high squared correlation coefficient with a low root-mean-square error, and they improve the MLR prediction model in 24%.