The applicability domain (AD) of a QSAR, does not receive as much attention as specific modeling methodologies or descriptor sets. We show that at least in terms of pharmaceutical applications, the AD should have high priority.In an unparalleled joint collaboration, Simulations Plus and the Bayer Pharma AG have developed a new in silico pKa prediction tool which is currently in use at Bayer and is a part of the release 7.0 of the Simulations Plus ADMET Predictor� software. The new model has been built with curated sets of ~14000 literature pKa values (~11000 compounds, representing literature chemical space) and ~19500 pKa values experimentally determined at Bayer Pharma (~16000 compounds, representing industry chemical space). We compare and quantify differences between the two domains. Model validation was performed with ~31000 new pKa values measured at Bayer. On the largest and most difficult test subset with >16000 pKa values (all from measurements on new compounds after training with the first Bayer training set) the old model achieved a mean absolute error (MAE) of 0.72, a root mean square error (RMSE) of 0.94, and a squared correlation coefficient (R2) of 0.87. The new model achieved significantly improved prediction statistics, with MAE=0.50, RMSE=0.67 and R2=0.93.At Bayer, the new pKa prediction model has been integrated into Pipeline Pilot and the PharmacophorInformatics platform PIx used by scientists at Bayer Pharma. Different output formats have been defined which allow customized use by Medicinal Chemists, Physical Chemists, and Computational Chemists. The fast calculation speed (>100,000 compounds/h [CPU: Intel Xeon L5420, 2.5 GHz]) allows Computational Chemists to automatically modify the protonation state of structures to reflect their dominant charge state(s) at a given pH (default 7.4) in order to prepare large libraries for high-throughput docking applications (e.g. virtual screening).

New tool for drug design and ADMET studies: Multiprotic pK<sub>a</sub> model with significantly improved prediction accuracy and vastly expanded applicability domain

247th National Meeting (2014)

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

New tool for drug design and ADMET studies: Multiprotic pKa model with significantly improved prediction accuracy and vastly expanded applicability domain

New tool for drug design and ADMET studies: Multiprotic pK_a model with significantly improved prediction accuracy and vastly expanded applicability domain