From Heuristic to Mathematical Modeling of Drugs Dissolution Profiles: Application of Artificial Neural Networks and Genetic Programming

Abstract

The purpose of this work was to develop a mathematical model of the drug dissolution (Q) from the solid lipid extrudates based on the empirical approach. Artificial neural networks (ANNs) and genetic programming (GP) tools were used. Sensitivity analysis of ANNs provided reduction of the original input vector. GP allowed creation of the mathematical equation in two major approaches: (1) direct modeling of Q versus extrudate diameter (d) and the time variable (t) and (2) indirect modeling through Weibull equation. ANNs provided also information about minimum achievable generalization error and the way to enhance the original dataset used for adjustment of the equations' parameters. Two inputs were found important for the drug dissolution: d and t. The extrudates length (L) was found not important. Both GP modeling approaches allowed creation of relatively simple equations with their predictive performance comparable to the ANNs (root mean squared error (RMSE) from 2.19 to 2.33). The direct mode of GP modeling of Q versus d and t resulted in the most robust model. The idea of how to combine ANNs and GP in order to escape ANNs' black-box drawback without losing their superior predictive performance was demonstrated. Open Source software was used to deliver the state-of-the-art models and modeling strategies.

Deep Learning in Drug Discovery

“…In fact, neural networks have a long and fruitful history in drug discovery and design. Because they bear the risk of being easily over‐trained and are perceived as a “black box”, they have often been substituted by other approaches such as SVM models 27,79. As noted by Winkler in a review article from 2004, continuous methodological advances in the field of neural networks alleviate some of the pitfalls and may have much to offer for hit and lead discovery 80.…”

Transparent computational intelligence models for pharmaceutical tableting process

“…The original data set contains six input features while the equation represents the two most important ones. This is an example of feature selection behaviour by rgp, which has been observed in other instances as well (Mendyk et al 2015). Feature selection densifies the effect of crucial inputs in the system and discards the trivial ones in an attempt to capture more information in the model yet making it simpler.…”

From Black-Box to Transparent Computational Intelligence Models: A Pharmaceutical Case Study

“…The original dataset contains six input features while the equation represents the three most important ones. This is an example of feature selection behavior by RGP, which has been observed in other instances as well [14]. Feature selection identifies the effect of crucial inputs in the system and discards the trivial ones in an attempt to capture more information in the model.…”