Bayesian Approach for Computing Free Energy on Perturbation Graphs with Cycles.

Area of Science:

• Computational Chemistry
• Statistical Mechanics
• Biophysics

Background:

• Calculating free energy differences between multiple states is crucial in molecular simulations.
• Perturbation graphs with cycles are commonly used, and the cycle consistency condition (zero free energy around cycles) can enhance accuracy.
• Existing methods may not fully leverage this condition.

Purpose of the Study:

• To develop a principled Bayesian method that couples free energy calculations across cycles.
• To improve the accuracy of free energy difference estimations by enforcing cycle consistency.
• To compare the new method against existing approaches.

Main Methods:

• Proposed the coupled Bayesian multistate Bennett acceptance ratio (CBayesMBAR) method.
• Applied CBayesMBAR to systems including harmonic oscillators and protein-ligand binding.
• Compared CBayesMBAR with methods ignoring cycle consistency and cycle closure correction.

Main Results:

• CBayesMBAR achieved more accurate free energy difference calculations.
• The method demonstrated superior performance for both harmonic oscillators and protein-ligand binding free energies.
• CBayesMBAR outperformed the cycle closure correction method.

Conclusions:

• The CBayesMBAR method effectively utilizes the cycle consistency condition for improved free energy calculations.
• This approach offers a more accurate and robust alternative for multistate free energy computations.
• CBayesMBAR advances the field of computational free energy calculations.