Toward Physics-Based Solubility Computation for Pharmaceuticals to Rival Informatics.

Area of Science:

• Computational chemistry
• Physical chemistry
• Drug discovery

Background:

• Accurate prediction of aqueous solubility is crucial for drug development.
• Existing physics-based methods often rely on approximations and limited sampling.
• Cheminformatics and machine learning offer alternative prediction strategies.

Purpose of the Study:

• To develop and validate an improved physics-based approach for calculating intrinsic aqueous solubility.
• To demonstrate that physics-based methods can achieve accuracy comparable to machine learning predictions.
• To extend the applicability of solubility predictions to systems not covered by informatics methods.

Main Methods:

• Utilized a sublimation thermodynamic cycle incorporating crystal phonon mode calculations.
• Employed the model-potential-based Ψmol method for lattice energy computations.
• Integrated hydration free energies from molecular dynamics or density functional theory simulations.

Main Results:

• Achieved accurate estimates of sublimation free energy by including vibrational contributions.
• Generated aqueous solubility predictions comparable to experimental and machine learning results.
• Demonstrated the method's adaptability for molecules with different conformational preferences across phases (coronene, succinic acid, desloratadine).

Conclusions:

• The enhanced physics-based model provides accurate aqueous solubility predictions.
• This approach offers a viable alternative to cheminformatics methods, especially for complex systems.
• The methodology is flexible and extensible for broader applications in computational chemistry and drug design.