A Data-Driven Dimensionality Reduction Approach to Compare and Classify Lipid Force Fields.

Area of Science:

• Computational biophysics and molecular modeling.
• Soft matter physics and materials science.

Background:

• Molecular dynamics (MD) simulations are vital for understanding lipid bilayer structural dynamics.
• Accurate comparison of classical force fields (FFs) for MD simulations remains a challenge.
• Existing FF assessments often rely on average properties, missing crucial local structural and dynamic differences.

Purpose of the Study:

• To develop an agnostic, high-dimensional method for comparing FFs at different resolutions (all-atom, united-atom, coarse-grained).
• To assess and classify 13 FFs for 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid bilayers.
• To capture non-average events and resolution-specific limitations of FFs.

Main Methods:

• Utilized Smooth Overlap of Atomic Position (SOAP) kernel for high-dimensional similarity metrics.
• Applied the SOAP-based metric to compare and classify 13 FFs modeling POPC bilayers.
• Analyzed local differences and non-average events, including phase transitions in dipalmitoylphosphatidylcholine (DPPC) bilayers.

Main Results:

• The SOAP kernel-based metric effectively compares, discriminates, and correlates FFs across different resolutions.
• The method captures subtle differences in modeling local structures and dynamics.
• Identified nucleation centers for the liquid-to-gel phase transition in DPPC bilayers, highlighting FF resolution limitations.

Conclusions:

• The proposed SOAP kernel metric offers an unbiased, high-dimensional approach for FF evaluation.
• This method provides deeper insights into FF performance beyond average properties.
• Reveals intrinsic resolution limitations of implicit versus explicit solvent FFs in modeling complex phenomena.