Accelerating molecular simulations of proteins using Bayesian inference on weak information.

Area of Science:

• Computational Biology
• Biophysics
• Structural Biology

Background:

• Atomistic molecular dynamics (MD) simulations are computationally intensive for predicting protein structures from amino acid sequences.
• Integrating heuristic or vague physical knowledge into MD simulations has been challenging.

Purpose of the Study:

• To develop a computationally efficient framework for protein structure prediction.
• To integrate "weak" external knowledge into molecular dynamics simulations for improved accuracy.

Main Methods:

• Developed a statistical mechanical framework, Modeling Using Limited Data with Coarse Physical Insight(s) (MELD + CPI).
• Integrated "weak" knowledge (e.g., "form a hydrophobic core") into atomistic replica-exchange molecular dynamics (REMD).

Main Results:

• MELD + CPI predicted native structures for 20 small proteins with an accuracy of <3.2 Å.
• Correctly identified native structures (<4 Å) for 15 out of 20 proteins, including ubiquitin.
• Achieved speeds up to five orders of magnitude faster than brute-force MD while satisfying detailed balance.

Conclusions:

• MELD + CPI offers a significant speedup for protein structure prediction.
• The framework effectively harnesses coarse physical insights for guiding simulations.
• This method shows promise for studying protein mechanisms and populations where limited physical knowledge is available.