Uniting Cheminformatics and Chemical Theory To Predict the Intrinsic Aqueous Solubility of Crystalline Druglike Molecules

Abstract

We present four models of solution free-energy prediction for druglike molecules utilizing cheminformatics descriptors and theoretically calculated thermodynamic values. We make predictions of solution free energy using physics-based theory alone and using machine learning/quantitative structure–property relationship (QSPR) models. We also develop machine learning models where the theoretical energies and cheminformatics descriptors are used as combined input. These models are used to predict solvation free energy. While direct theoretical calculation does not give accurate results in this approach, machine learning is able to give predictions with a root mean squared error (RMSE) of ∼1.1 log S units in a 10-fold cross-validation for our Drug-Like-Solubility-100 (DLS-100) dataset of 100 druglike molecules. We find that a model built using energy terms from our theoretical methodology as descriptors is marginally less predictive than one built on Chemistry Development Kit (CDK) descriptors. Combining both sets of descriptors allows a further but very modest improvement in the predictions. However, in some cases, this is a statistically significant enhancement. These results suggest that there is little complementarity between the chemical information provided by these two sets of descriptors, despite their different sources and methods of calculation. Our machine learning models are also able to predict the well-known Solubility Challenge dataset with an RMSE value of 0.9–1.0 log S units.

Are the Sublimation Thermodynamics of Organic Molecules Predictable?

“…These methods have been previously discussed in a number of publications 10,19,27,106 and are only briefly outlined below. The workflow is outlined in Scheme 1.…”

“…It has previously been reported that for solubility, combining descriptors from theoretical chemistry with 2D cheminformatics descriptors does not notably improve the model (the descriptor sets were non-complementary). 10 However, for sublimation thermodynamics, it seems that the two descriptor sets are complementary. A reduction in prediction RMSE is accompanied by an improvement in R 2 for the models that combine the two descriptor sets.…”

“…Many of these solvation/hydration methodologies have been summarized in a recent review by Skyner et al 19 Commonly, QSPR methods are employed due to their speed, convenience and accuracy, when provided with a suitable training dataset. 10,[20][21][22]23 These methods represent the current state-of-the-art in practical solubility prediction. However, QSPR methods lack the theoretical basis of a fundamental physical theory, hence limiting their interpretability and the understanding that can be gained from their use.…”

Hydration Free Energy as a Molecular Descriptor in Drug Design: A Feasibility Study

“…[13,14] The logarithm of solubility (log S) is relatedt oDG hyd as shown in Equations 1a nd 2. [15] (R = the gas constant and T = temperature) Log S ¼À DG solution 2:303RT ð1Þ…”

Suggested Improvements for Measurement of Equilibrium Solubility-pH of Ionizable Drugs

“…Since the mid-1990s there has been a heightened effort in drug discovery to predict drug-relevant aqueous solubility, described in at least a hundred publications (e.g., Huuskonen et al [1,2]; Abraham and Le [3]; Jorgensen and Duffy [4,5]; Bergström et al [6]; Hou et al [7]; Delaney [8]; Dearden [9]; Balakin et al [10]; Taskinen and Norinder [11]; Jain and Yalkowsky [12]; Shayanfar and Jouyban [13]; Wang and Hou [14]; Elder and Holm [15]; McDonagh et al [16]). The typical errors in drug solubility prediction are 0.7 -1.0 log unit, and for low-soluble compounds, errors are considerably greater than a log unit (Jorgensen and Duffy [5]; Palmer and Mitchell [17]).…”