pKa directly affects what tautomers and charge states a molecule acquires in solution, knowledge of which is key to drug design projects for assessing ADME properties like logP/D, membrane permeability, solubility, and protein binding, as well as for generating inputs for virtual screening. ML-Epik is a program for predicting the pKa values as well as the tautomeric and charge state distribution of drug-like molecules in an aqueous solution. ML-Epik predicts pKa values using an ensemble of atomic graph convolutional neural network models and then calculates state populations using the predicted pKa values for each microstate. In general, ML-Epik can predict pKa values to within ~0.5 pKa units median absolute deviation, comparable to the best available empirical and semiempirical methods. ML-Epik is fast and due to the nature of ML technology and the extensive training data set, has broad coverage for the types of novel chemistries explored by medicinal chemists.

ML-Epik: Rapid pKa and microstate distribution prediction by machine learning

ACS Spring Meeting 2021

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