Prediction and assessment of pharmacokinetic (PK) properties (ADME--absorption, distribution, metabolism and excretion) and toxicity (T) of investigative compounds plays a critical role in the early stages of drug discovery and design. Recent advancements in AI have made it possible to use in silico approaches to evaluate and optimize the ADME/T properties for drugs leading to the development of an effective and safe drug candidate at an enterprise scale. To this end, we have developed an open-source command-line interface (CLI) platform using autoML algorithms from PyCaret and H2O framework to build quantitative structure-activity relationship (QSAR) and quantitative structure-property relationships (QSPR) models based on molecular descriptors and fingerprints as attributes. We have used publicly available large datasets of structurally diverse compounds to build predictive models for four endpoints: (1) plasma protein binding (PPB), (2) aqueous solubility, (3) microsomal stability, and (4) hERG K+ channel blockage with accuracies better than or comparable to any models published in the literature. This streamlined drug discovery platform provides the user with an end-to-end automated workflow starting with data processing, feature ranking, model construction and model selection from an ensemble of models. While the current study has only focused on four endpoints, the workflow is easily generalizable to any endpoint of importance and can serve as a benchmark for users to assess and optimize the physicochemical and PK profiles of the chemical compounds in pharmaceutical R&D projects.

Scaling AI to develop QSAR/QSPR classification and regression models for predictive analysis of pharmacokinetic and toxicity endpoints

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.