Purely Predicting the Pharmaceutical Solubility: What to Expect from PC-SAFT and COSMO-RS?

Area of Science:

• Computational chemistry and thermodynamics
• Pharmaceutical sciences
• Physical chemistry

Background:

• Accurate prediction of active pharmaceutical ingredient (API) solubility is crucial for drug development.
• Thermodynamic models like PC-SAFT and COSMO-RS are used for solubility predictions.
• These models differ in their reliance on experimental data for parameterization.

Purpose of the Study:

• To benchmark the predictive performance of PC-SAFT and COSMO-RS for API solubility in pure solvents.
• To evaluate these models in a purely predictive regime, minimizing reliance on experimental solubility data.
• To compare the qualitative and quantitative accuracy of both models for API solubility.

Main Methods:

• Benchmarking PC-SAFT and COSMO-RS for predicting the solubility of 10 APIs in various pure solvents.
• Utilizing a fair comparison by omitting binary interaction parameters for PC-SAFT (k=0) and using API parameters not trained on solubility data.
• Assessing predictions against a large experimental dataset, focusing on activity coefficients and fusion properties.

Main Results:

• COSMO-RS significantly outperformed PC-SAFT in both qualitative (solvent ranking) and quantitative predictions.
• COSMO-RS showed superior performance for 9 out of 10 APIs and 63% of API-solvent systems.
• Root-mean-square deviations were 0.82 log units for COSMO-RS and 1.44 log units for PC-SAFT.

Conclusions:

• COSMO-RS offers a more reliable, purely predictive approach for API solubility compared to PC-SAFT.
• Both models exhibited limitations, with frequent incorrect qualitative predictions of deviations from ideality.
• The study highlights the sensitivity of solubility predictions to API parametrization in both models.