Accelerating physical simulations of proteins by leveraging external knowledge.

Area of Science:

• Computational Biology
• Molecular Physics
• Biophysics

Background:

• Calculating protein structure-function relationships is computationally intensive due to the vast conformational space.
• Existing methods like Monte Carlo and molecular dynamics (MD) struggle with exponential scaling challenges.
• Explore-and-Exploit (EE) methods are widely used for optimization but have limitations.

Purpose of the Study:

• To introduce MELD, a Bayesian method designed to enhance computational efficiency in molecular modeling.
• To demonstrate MELD's ability to integrate diverse external information into simulation processes.
• To accelerate the computation of protein structure-function relationships.

Main Methods:

• Developed a Bayesian approach named MELD (Model-based Exploration and Learning with Data).
• MELD integrates 'explore-and-exploit' strategies with external information sources (vague, noisy, heuristic, experimental data).
• Applied MELD to accelerate physical MD simulations for protein structure determination and binding affinity prediction.

Main Results:

• MELD significantly accelerates physical MD simulations for protein structure determination using experimental data.
• The method improves protein structure prediction accuracy when guided by heuristic directives.
• MELD enhances the prediction of protein binding affinities with limited binding site information.

Conclusions:

• MELD offers a novel 'Guided Explore-and-Exploit' approach for molecular modeling.
• This Bayesian method effectively integrates external data to overcome computational scaling issues.
• The MELD framework shows potential applicability beyond molecular science, in diverse computational fields.