# Comparison of Sampling Methods via Robust Free Energy Inference: Application to Calmodulin

## TL;DR

This paper introduces a Bayesian free energy estimation method to compare sampling techniques in molecular dynamics, demonstrating its robustness and effectiveness in different calmodulin states.

## Contribution

A novel Bayesian free energy inference method (GM method) is proposed, offering improved robustness and predictive capabilities over existing estimators for molecular sampling analysis.

## Key findings

- Regular MD is most efficient for diffusion-dominated apo-calmodulin.
- Enhanced sampling methods outperform in rugged holo-calmodulin landscapes.
- The GM method provides robust free energy estimates with sparse data.

## Abstract

A free energy landscape estimation-method based on Bayesian inference is presented and used for comparing the efficiency of thermally enhanced sampling methods with respect to regular molecular dynamics, where the simulations are carried out on two binding states of calmodulin. The proposed free energy estimation method (the GM method) is compared to other estimators using a toy model showing that the GM method provides a robust estimate not subject to overfitting. The continuous nature of the GM method, as well as predictive inference on the number of basis functions, provide better estimates on sparse data. We find that the free energy diffusion proper- ties determine sampling method effectiveness, such that the diffusion dominated apo-calmodulin is most efficiently sampled by regular molecular dynamics, while the holo with its rugged free energy landscape is better sampled by enhanced methods.