Free Energy Perturbation Approach for Accurate Crystalline Aqueous Solubility Predictions.

Area of Science:

• Computational chemistry
• Drug discovery and development

Background:

• Accurate prediction of crystalline thermodynamic solubility is crucial for drug discovery, particularly for poorly soluble compounds.
• Current methods struggle with the early assessment of solubility, hindering the development of new drug candidates.

Purpose of the Study:

• To evaluate a physics-based Free Energy Perturbation (FEP+) approach for computing thermodynamic aqueous solubility.
• To assess the predictive and differentiating power of FEP+ across diverse chemical spaces.
• To compare FEP+ performance against state-of-the-art machine learning approaches.

Main Methods:

• Detailed evaluation of the Free Energy Perturbation (FEP+) method.
• Application of FEP+ to pharmaceutically relevant literature compounds and complex AbbVie compounds.
• Comparison of FEP+ predictions with experimental solubility data and machine learning models.

Main Results:

• The FEP+ approach achieved a root-mean-square error (RMSE) of 0.86 and an R-squared (R²) of 0.69.
• FEP+ demonstrated superior predictive and differentiating power compared to machine learning approaches using quantum mechanics-based descriptors.
• Explicit consideration of crystalline packing was found to be important for accurate solubility prediction using FEP+.

Conclusions:

• The physics-based FEP+ approach offers a powerful tool for predicting thermodynamic aqueous solubility.
• FEP+ provides improved correlations to experimental solubility, aiding in the selection of drug candidates.
• Computed energetics from FEP+ can offer valuable insights for molecule design within the drug discovery cycle.