Principled Approach for Computing Free Energy on Perturbation Graphs with Cycles

Xinqiang Ding,John Drohan

DOI: https://doi.org/10.26434/chemrxiv-2024-nkjwc

2024-07-22

Abstract:A common approach for computing free energy differences among multiple states is to build a perturbation graph connecting the states and compute free energy differences on all edges of the graph. Such perturbation graphs are often designed to have cycles. Because free energy is a function of states, the free energy around any cycle is zero, which we refer to as the cycle consistency condition. Since the cycle consistency condition relates free energy differences on edges of a cycle, it could be used to improve the accuracy of free energy estimates. Here we propose a Bayesian method called coupled Bayesian multistate Bennett acceptance ratio (CBayesMBAR) that can properly couple the calculations of free energy differences on edges of cycles in a principled way. We apply CBayesMBAR to compute free energy differences among harmonic oscillators and relative protein-ligand binding free energies. In both cases, CBayesMBAR provides more accurate results compared to methods that do not consider the cycle consistency condition. Additionally, it outperforms the cycle closure correction method that also uses cycle consistency conditions.

Chemistry

What problem does this paper attempt to address?

The paper aims to address the problem of calculating free energy differences between multiple thermodynamic states in computational chemistry, particularly when these states are connected through a perturbation graph that contains cycles, and how to improve the accuracy of free energy estimates in such scenarios.

The authors propose a new algorithm based on Bayesian methods—Coupled Bayesian Multistate Bennett Acceptance Ratio (CBayesMBAR). This algorithm effectively utilizes the cycle consistency condition to improve the accuracy of free energy difference estimates. The cycle consistency condition states that the free energy difference between states in a cycle should be independent of the chosen path, meaning the total free energy change within the cycle is zero.

Specifically, the key aspects of the CBayesMBAR method are:
1. **Incorporating Cycle Consistency Condition**: Encoding the cycle consistency condition into the prior distribution to ensure that the free energy estimates directly satisfy this condition.
2. **Coupled Computation**: Using a Bayesian framework to couple the free energy calculations across multiple states, thereby leveraging high-precision edge information to improve the estimates of low-precision edges.
3. **Uncertainty Quantification**: Quantifying the uncertainty of free energy estimates using the standard deviation of the posterior distribution, which takes into account both sampling configurations and the cycle consistency condition.

The effectiveness of CBayesMBAR is validated through two example applications:
1. **Harmonic Oscillator System**: In a system composed of four 2D harmonic oscillators, CBayesMBAR significantly improved the accuracy of free energy difference estimates, outperforming independent BayesMBAR calculations and the Cycle Closure Correction (CCC) method.
2. **Relative Protein-Ligand Binding Free Energy**: In a system involving six ligands interacting with the Tyk2 protein, CBayesMBAR also provided more accurate free energy difference estimates compared to BayesMBAR and CCC.

In summary, CBayesMBAR demonstrates both theoretically and experimentally its advantage in improving the accuracy of free energy difference estimates by effectively utilizing the cycle consistency condition in perturbation graphs.