Conformer Generation Workflows for COSMO-RS Calculations: Are They All the Same?

Area of Science:

• Computational chemistry
• Drug discovery and development
• Physical chemistry

Background:

• Accurate prediction of drug solubility is crucial for pharmaceutical formulation and bioavailability.
• The Conductor-like Screening Model for Real Solvents (COSMO-RS) is a key tool for estimating solubility, but its accuracy depends heavily on input molecular conformations.
• Previous studies lacked a systematic evaluation of different conformer generation methods' impact on COSMO-RS solubility predictions across diverse datasets.

Purpose of the Study:

• To systematically evaluate the influence of various conformer generation workflows on the accuracy of COSMO-RS solubility predictions.
• To compare the performance of different computational approaches, from simple to advanced, using a large, curated dataset.
• To assess the benefits of using conformer ensembles versus single-conformer pipelines and explore solvent-aware methods.

Main Methods:

• Assessed four distinct conformer generation workflows: RDKit-based, semiempirical, and DFT-based.
• Utilized a curated dataset of 926 experimental solubility values and 405 free energies of solvation.
• Compared conformer ensembles against solvent-aware single-conformer pipelines for selected molecules.

Main Results:

• All evaluated conformer generation workflows demonstrated comparable accuracy, with mean absolute errors around one logarithmic unit.
• A consistent underperformance was observed in the low-solubility regime across all methods, highlighting a persistent challenge.
• Conformer ensembles offered moderate improvements but did not consistently outperform solvent-aware lowest-energy conformer selection.
• Incorporating solvent information during lowest-energy conformer identification enhanced accuracy for molecules with intramolecular interactions.

Conclusions:

• Practical guidance is provided for selecting conformer generation strategies in high-throughput COSMO-RS applications.
• Solvent-aware workflows are identified as a promising avenue for improving predictive accuracy and robustness without sacrificing computational efficiency.
• Further research into optimizing conformer generation, especially for low-solubility compounds, is warranted.