Learning Effective Molecular Models from Experimental Observables.

Area of Science:

• Computational biology
• Biophysics
• Molecular dynamics

Background:

• Coarse-grained (CG) models offer a computational advantage for simulating large macromolecules over long timescales.
• However, CG models involve approximations that can limit their predictive accuracy.
• Atomistic simulations are computationally expensive for large systems and long timescales.

Purpose of the Study:

• To develop a framework for designing accurate CG models that reproduce experimental observables.
• To validate the modeling framework using the folding mechanism of a WW domain protein.
• To assess the ability of optimized CG models to predict properties not explicitly targeted during optimization.

Main Methods:

• Development of a novel modeling framework for CG model design.
• Application of the framework to simulate the folding mechanism of a WW domain.
• Optimization of CG models to match reference simulation data and experimental observables.
• Analysis of protein dynamics and identification of conserved features.

Main Results:

• Optimized CG models accurately reproduced experimental data and reference simulations for WW domain folding.
• The models successfully predicted additional observables not directly included in the optimization targets.
• Localized frustration was identified as a key factor in the protein's folding mechanism.
• Evidence suggests evolutionary conservation of specific protein dynamics.

Conclusions:

• The proposed framework enables the design of accurate and predictive CG models for macromolecular dynamics.
• Coarse-graining resolution is critical for capturing essential biological features.
• Localized frustration and conserved dynamics are important aspects of protein folding mechanisms.