Computational models to predict the binding of small molecules to Cytochrome P450 2D6 to replace the use of animals
Abstract: The superfamily of enzymes known as the Cytochromes P450 (CYPs), are involved in the metabolism of a large number of endogenous and exogenous compounds. This toxicological role means that CYP enzymes are a primary consideration for early stage drug design. This has led to a high interest in the ability to use computational models to accurately predict CYP inhibition and facilitate a “fail-fast” methodology for any compounds with toxicological liabilities. There is also the potential for such technology to replace and/or minimize the use of animal models for determining CYP inhibition. It is of particular interest to be able to predict the interaction of compounds with CYP2D6, due to the polymorphic nature of this enzyme.
This project focuses on several aspects of the Quantitative Structure Activity Relationship (QSAR) modeling of CYPs and how to increase the quality of such models. A range of QSAR models were built and analyzed to establish traditional descriptor-based models for the prediction of CYP2D6 inhibition, alongside QSAR models based on descriptors derived solely from the .mol2 atom types of compounds in a small, self-consistent experimental dataset.
Preliminary analysis of the dataset identified the need to split it into two smaller subsets – termed the diverse subset and the analogue subset. The results of the experiments show that the best QSAR equation for the diverse subset was a 5- descriptor model. The r2 value for that model is 0.62 and the q2 is 0.53 which shows reasonable levels of predictivity and robustness. The second model, based on the analogue subset also gave promising results with an r2 value of 0.66 and q2 of 0.47. Initial models based on atom-types alone showed that the models were less predictive than descriptor based models. It was postulated that this was due to an explicit representation of the electronic characteristics of the molecules in those models.