Broad chemical transferability in structure-based coarse-graining.

Area of Science:

• Computational chemistry
• Materials science
• Molecular modeling

Background:

• Bottom-up coarse-grained (CG) models offer higher structural fidelity than top-down models due to their link to higher-resolution references, making them chemically specific.
• However, this chemical specificity can hinder transferability, which is crucial for compound-screening strategies.

Purpose of the Study:

• To reconcile bottom-up, structure-preserving CG models with chemical transferability.
• To develop a CG parameterization approach that is both structure-based and chemically transferable.

Main Methods:

• Utilized atomic representations, unsupervised learning, and large-scale extended-ensemble force-matching parameterization for 3441 C7O2 small-molecule isomers.
• Identified 19 representative molecules encoding the local environment of gas-phase conformers.
• Obtained reference interactions from bulk liquids and binary mixtures across 703 state points.

Main Results:

• Developed a CG model that is both structure-based and chemically transferable.
• The resulting force field demonstrated, on average, higher structural accuracy than single-state-point equivalents.
• Extended-ensemble averaging acted as a mean-force regularizer, smoothing overly specific force and structural correlations.

Conclusions:

• The presented approach successfully reconciles bottom-up CG model fidelity with chemical transferability.
• The developed CG bead types facilitate the construction of new molecules while maintaining structure-based parameterization benefits.
• This method enhances the utility of CG models for compound screening and molecular design.