On the removal of initial state bias from simulation data.

Area of Science:

• Computational biology
• Molecular dynamics simulations
• Biomolecular modeling

Background:

• Atomistic simulations are crucial in molecular life science.
• Current computing architectures favor parallel trajectory methods.
• Advanced sampling strategies often introduce bias by not using correct Boltzmann weights.

Purpose of the Study:

• Analyze the performance of Markov state models (MSMs) for thermodynamic reweighting.
• Evaluate MSMs' ability to correct biases from adaptive sampling methods.
• Compare MSMs with alternative methods like statistical resampling.

Main Methods:

• Applied MSMs to a hierarchical set of systems.
• Investigated thermodynamic reweighting capabilities of MSMs.
• Utilized pure likelihood-based inference for transition matrix estimation.
• Compared MSM performance against statistical resampling.

Main Results:

• MSMs can rigorously recover equilibrium distributions for low-dimensional systems if local flux imbalances are preserved.
• Pure likelihood-based inference yielded the best results for a real-world biomolecular system.
• Bias removal by MSMs was incomplete in the tested real-world application.
• Statistical resampling outperformed tested MSMs for the real-world system.

Conclusions:

• MSMs are powerful for thermodynamic reweighting in low-dimensional systems.
• Careful implementation is needed to avoid introducing or failing to correct biases.
• Alternative methods like statistical resampling may be more effective for complex, real-world systems.
• Recommendations are provided for addressing reweighting challenges in practice.