SAMPL6 challenge results from $$pK_a$$ predictions based on a general Gaussian process model

Abstract

A variety of fields would benefit from accurate pKa predictions, especially drug design due to the effect a change in ionization state can have on a molecule’s physiochemical properties. Participants in the recent SAMPL6 blind challenge were asked to submit predictions for microscopic and macroscopic pKas of 24 drug like small molecules. We recently built a general model for predicting pKas using a Gaussian process regression trained using physical and chemical features of each ionizable group. Our pipeline takes a molecular graph and uses the OpenEye Toolkits to calculate features describing the removal of a proton. These features are fed into a Scikit-learn Gaussian process to predict microscopic pKas which are then used to analytically determine macroscopic pKas. Our Gaussian process is trained on a set of 2,700 macroscopic pKas from monoprotic and select diprotic molecules. Here, we share our results for microscopic and macroscopic predictions in the SAMPL6 challenge. Overall, we ranked in the middle of the pack compared to other participants, but our fairly good agreement with experiment is still promising considering the challenge molecules are chemically diverse and often polyprotic while our training set is predominately monoprotic. Of particular importance to us when building this model was to include an uncertainty estimate based on the chemistry of the molecule that would reflect the likely accuracy of our prediction. Our model reports large uncertainties for the molecules that appear to have chemistry outside our domain of applicability, along with good agreement in quantile-quantile plots, indicating it can predict its own accuracy. The challenge highlighted a variety of means to improve our model, including adding more polyprotic molecules to our training set and more carefully considering what functional groups we do or do not identify as ionizable.

Holistic Prediction of the pKa in Diverse Solvents Based on a Machine‐Learning Approach

“…The submitted prediction strategies included quantum-chemistry based calculation, [19] EC-RISM theory, [20] QM/MM approach, [21] ab initio quantum mechanical prediction [22] as well as machine learning. [23] Them achine learning methods were built with the general Gaussian process with unfortunately only moderate accuracy.B ased on the HM-XGBoost model, we obtained aprediction with MAE = 0.80 and RMSE = 1.07 (Figure 9and Figure S17 A). Theprediction with holistic NN model also gave comparable results (Figure S17 B), with slightly higher MAE and RMSE.…”

MLatom: A program package for quantum chemical research assisted by machine learning

“…The Matérn kernel has several interesting properties which makes it an increasingly popular choice . First, it reduces to the exponential kernel for n = 0: which is not implemented separately in MLatom .…”

“…[23,24] The Matérn kernel has several interesting properties which makes it an increasingly popular choice. [14,[25][26][27][28][29][30][31][32][33] First, it reduces to the exponential kernel for n = 0: [24]…”

SAMPL6 Octanol–water partition coefficients from alchemical free energy calculations with MBIS atomic charges

“…In the past eleven years SAMPL challenges have included blind prediction of a variety of different properties such as hydration free energy [1], binding affinity of host-guest systems [2,3], distribution coefficients [4,5] and calculations of pka [6]. They have made an important contribution to the development of new methods and computational tools [7] and increased the accuracy in the prediction of each of these properties.…”

“…The challenge of SAMPL6 part II consisted in the determination of octanol water partition coefficients of 11 molecules (see Fig. 1) that are similar to fragments of protein kinase inhibitors and are a subset of molecules that were part of the pka SAMPL6 challenge [6]. Predicting the logarithm of the partition coefficient is challenging both experimentally [8] and computationally, and in general there are not many reports in which different computational methodologies are tested blindly (without knowledge of the experimental results) and then compared to each other.…”