MDSubSampler

MDSubSampler performs a posteriori subsampling of biomolecular molecular dynamics (MD) trajectories—including atomistic and coarse-grained simulations at microsecond timescales—to reduce dataset size while preserving distributions of relevant geometrical properties for downstream analysis of protein conformations and dynamics.

Key Features:

• Python library: Implements programmatic subsampling methods for post-processing MD simulation data.
• Uniform Sampling: Ensures equal representation across the dataset by selecting frames to achieve unbiased coverage.
• Random Sampling: Performs stochastic selection of frames for exploratory or unbiased subsets.
• Stratified Sampling: Divides datasets into strata and samples from each to preserve subgroup characteristics.
• Weighted Sampling: Assigns selection probabilities to frames based on predefined criteria to emphasize specific features or conformations.
• Bootstrapping Sampling: Employs resampling with replacement to estimate statistical properties and assess variability.
• Distribution preservation: Performs sampling while preserving the original distribution of relevant geometrical properties within the dataset.
• GPU-accelerated trajectory support: Operates on trajectories produced by GPU-based MD algorithms, including long microsecond-scale simulations.

Scientific Applications:

• Noise reduction: Reduces redundancy and noise in large MD trajectory datasets through targeted subsampling.
• Structure selection for ensemble docking: Selects representative protein conformations for ensemble docking and structure-based studies.
• Statistical estimation: Enables estimation of statistical properties and assessment of variability via bootstrapping for downstream analyses.
• Post-processing of large-scale simulations: Facilitates downstream analyses by producing smaller, representative subsets from extensive simulation outputs.

Methodology:

Performs a posteriori subsampling using uniform, random, stratified, weighted, and bootstrapping (resampling with replacement) methods applied to MD trajectories to preserve distributions of relevant geometrical properties.