Resolution limit of data-driven coarse-grained models spanning chemical space

Area of Science:

• Computational chemistry
• Materials science
• Molecular modeling

Background:

• Materials design faces challenges due to the vast chemical compound space (CCS).
• Chemically transferable coarse-grained models improve screening efficiency by mapping diverse molecular fragments to single bead types.

Purpose of the Study:

• To propose new criteria for designing transferable coarse-grained models.
• To evaluate the Martini model within this new framework and explore transferability.
• To investigate the impact of varying bead types on model accuracy and chemical mapping.

Main Methods:

• Parameterizing three Martini-like coarse-grained models with 5 to 16 bead types.
• Interpolating the Martini interaction matrix for force field compatibility.
• Employing a Bayesian approach to map bead types to chemical groups.

Main Results:

• Comparable accuracy to Martini achieved with fewer bead types, using water/octanol partitioning free energy (ΔG_W→Ol) as a metric.
• Increased number of bead types reduces uncertainty in back-mapping bead types to specific chemistries.
• A resolution limit exists when using only ΔG_W→Ol for coarse-graining CCS due to large fragment mapping.

Conclusions:

• The proposed criteria enhance the design of transferable coarse-grained models for materials discovery.
• Fewer bead types can yield comparable accuracy, while more bead types improve chemical specificity.
• Further refinement is needed to overcome resolution limitations in coarse-grained models for CCS exploration.