Machine Learning-Guided Approach for Studying Solvation Environments.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Chemical Engineering

Background:

• Understanding solvation environments at the molecular level is crucial in chemistry, biology, and engineering.
• Accurate modeling of local solvation effects for diverse solutes and solvents remains a challenge.

Purpose of the Study:

• To develop a practical, all-quantum mechanics-based approach for calculating single-ion solvation free energies.
• To model local solvation effects for any solute in any solvent without requiring dynamics simulations.

Main Methods:

• A static, cluster-continuum approach utilizing global optimization to identify low-energy molecular clusters.
• Employing the smooth overlap for atomic positions (SOAP) learning kernel to quantify solute environment similarity.
• Utilizing sketch maps, a nonlinear dimensionality reduction algorithm, for visualizing solute environment similarity.

Main Results:

• The approach was tested on ions with charges 2+, 1+, 1-, and 2-.
• A correlation was observed between the solvation environment's similarity and the calculated single-ion solvation free energy.
• Calculated solvation free energies were within 5% of experimental measurements for most tested ions.

Conclusions:

• The developed method provides accurate single-ion solvation free energies without dynamics simulations or prior knowledge of solvation structure.
• This approach is transferable to other systems where dynamics simulations are computationally intensive or not feasible.