Predicting intrinsic aqueous solubility by a thermodynamic cycle.

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Accurate prediction of intrinsic aqueous solubility is essential for drug development.
• Direct ab initio calculations of thermodynamic properties often lack the required accuracy for solubility prediction.
• Existing methods may require knowledge of specific crystalline forms, limiting applicability.

Purpose of the Study:

• To develop a reliable computational method for predicting the intrinsic aqueous solubility of crystalline organic molecules.
• To improve the accuracy of solubility predictions beyond direct ab initio calculations.
• To create a model applicable even when the polymorphic form is unknown.

Main Methods:

• Utilized two different thermodynamic cycles for solubility calculations.
• Combined direct computation of thermodynamic properties with informatics-based regression models.
• Incorporated calculated crystal lattice energy as a key descriptor in the regression model.

Main Results:

• A three-variable linear regression model achieved high accuracy for predicting log intrinsic solubility (mol/L).
• The model demonstrated good performance on an external test set of drug molecules, with r(2)=0.77 and RMSE=0.71.
• The inclusion of crystal lattice energy effectively accounted for solid-state interactions.

Conclusions:

• A hybrid computational and informatics approach provides accurate predictions of intrinsic aqueous solubility.
• The developed method does not require prior knowledge of the crystalline polymorphic form.
• The approach shows potential for application to other solid-state systems like salts and cocrystals.