Scaling and Merging Time-Resolved Laue Data with Variational Inference.

Area of Science:

• Structural biology
• Biophysics
• Computational chemistry

Background:

• Time-resolved X-ray crystallography (TR-X) enables atomic-resolution studies of molecular dynamics.
• Analyzing TR-X data is challenging due to difficulties in extracting small, time-dependent signals.
• Variational inference (VI) has shown promise in merging redundant observations to overcome these analysis bottlenecks.

Purpose of the Study:

• To present a successful application of variational inference (VI) for analyzing time-resolved X-ray diffraction data.
• To demonstrate a strategy for extracting high signal-to-noise electron density changes from TR-X data.
• To provide a practical example for researchers using VI in time-resolved crystallography.

Main Methods:

• Application of variational inference (VI) to time-resolved X-ray crystallography data.
• Analysis of time-dependent structural changes in the enzyme DJ-1 upon substrate binding.
• Ablation study to systematically evaluate the impact of hyperparameter choices on model performance.

Main Results:

• Successful extraction of high signal-to-noise electron density changes from DJ-1 TR-X data using VI.
• Demonstration of VI's effectiveness in identifying subtle molecular dynamics.
• Quantification of the influence of individual hyperparameters on the VI model's success.

Conclusions:

• Variational inference provides a robust statistical framework for analyzing challenging time-resolved X-ray crystallography data.
• The presented strategy and ablation study offer valuable insights for optimizing VI hyperparameter selection.
• This work serves as a practical guide for applying VI to advance molecular dynamics studies in structural biology.