Best of Both Worlds: Combining Pharma Data and State of the Art Modeling Technology To Improve in Silico pK_a Prediction

Abstract

In a unique collaboration between a software company and a pharmaceutical company, we were able to develop a new in silico pKa prediction tool with outstanding prediction quality. An existing pKa prediction method from Simulations Plus based on artificial neural network ensembles (ANNE), microstates analysis, and literature data was retrained with a large homogeneous data set of drug-like molecules from Bayer. The new model was thus built with curated sets of ∼14,000 literature pKa values (∼11,000 compounds, representing literature chemical space) and ∼19,500 pKa values experimentally determined at Bayer Pharma (∼16,000 compounds, representing industry chemical space). Model validation was performed with several test sets consisting of a total of ∼31,000 new pKa values measured at Bayer. For the largest and most difficult test set with >16,000 pKa values that were not used for training, the original model achieved a mean absolute error (MAE) of 0.72, root-mean-square error (RMSE) of 0.94, and squared correlation coefficient (R(2)) of 0.87. The new model achieves significantly improved prediction statistics, with MAE = 0.50, RMSE = 0.67, and R(2) = 0.93. It is commercially available as part of the Simulations Plus ADMET Predictor release 7.0. Good predictions are only of value when delivered effectively to those who can use them. The new pKa prediction model has been integrated into Pipeline Pilot and the PharmacophorInformatics (PIx) platform used by scientists at Bayer Pharma. Different output formats allow customized application by medicinal chemists, physical chemists, and computational chemists.

Benefit of Retraining pKa Models Studied Using Internally Measured Data

“…They showed that training can reduce the average prediction error for new chemotypes to what is observed for public domain data. A similar study was published by Fraczkiewicz et al 18 for the pKa prediction method from Simulations Plus. The initial model was trained on about 14,000 literature pKa values.…”

“…18,34 As the MoKa software is not differentiating between macro-and microconstants in the training, we need to consider this problem in our curation effort by not blindly assigning experimental pKa values to ionization sites. Instead we restrict the assignment to cases where we can unambiguously make it and where the macroconstant is a good approximation of the microconstant.…”

Every Jack has His Jill: Finding a Target for Your Combinatorial Library

“…First, protonation states at pH 7.4 for co-crystallized ligand, ChEMBL hits and library compounds were calculated using the pKa module co-developed by Bayer and SimulationPlus [31] and implemented as Pipeline Pilot component "ADMET predictor" [32], while ring conformers, tautomers and stereoisomers were generated using Schrödinger LigPrep utility, release 9.8.…”

“…Amino acids with different side chain orientations responsible for change in pocket shape are shown as sticks. SiteMap [31] generated surfaces are shown in blue mesh for 3G1M and magenta mesh for 3O8H; b) Overlay of the crystallized 3G1M ligand (green), the crystallized_3O8H ligand (cyan) and the docking pose of the 3O8H ligand in 3G1M (magenta). for our designed oxadiazole library.…”

Tales from the war on error: the art and science of curating QSAR data

“…Many of the errors described were discovered (and corrected) while curating data sets used in constructing some of the many models distributed as part of ADMET Predictor TM [7]: our pK a model [8]; cytochrome P450 (CYP) site-of-metabolism models for CYPs 1A2, 2A6, 2B6, 2C8, 2C9, 2C19, 2D6, 2E1 and 3A4 and their associated substrate classification models; and kinetic models (inhibition, K m , V max and intrinsic clearance (CL int )) for the five major CYPs-1A2, 2C9, 2C19, 2D6 and 3A4. Others were discovered in the course of selecting a reference subset of the World Drug Index (WDI) [9] to set thresholds of concern (ADMET Risk TM thresholds) for 40 other ADMET models currently included in ADMET Predictor's ADMET Risk TM score.…”