Robust Automated Truncation Point Selection for Molecular Simulations.

Area of Science:

• Computational chemistry
• Statistical mechanics
• Molecular dynamics simulations

Background:

• Molecular simulations often suffer from initial bias due to unrepresentative starting configurations.
• Discarding initial data is a common practice to mitigate this bias.
• Chodera's method is a popular automated approach for selecting truncation points, but requires further assessment.

Purpose of the Study:

• To reformulate White's marginal standard error rule for truncation point selection.
• To develop and evaluate a spectrum of heuristics that account for autocorrelation.
• To compare these new methods against Chodera's method using synthetic time series data.

Main Methods:

• Reformulation of White's marginal standard error rule.
• Development of truncation point selection heuristics with varying autocorrelation treatment.
• Testing on synthetic time series data mimicking free energy calculations.
• Implementation in the open-source Python package RED.

Main Results:

• Methods with thorough autocorrelation consideration showed late, variable truncation.
• Methods with less autocorrelation consideration resulted in early truncation, increasing bias.
• A recommended method balances these extremes for robust performance.
• None of the tested methods reliably detected insufficient sampling.

Conclusions:

• A new spectrum of truncation point selection heuristics offers improved performance over existing methods.
• Balancing autocorrelation treatment is key for robust molecular simulation data analysis.
• Further development is needed for methods that reliably detect insufficient sampling.