TWIGS

TWIGS identifies modules of coordinated activity in three-way biological datasets (e.g., patient-gene-time or subject-voxel-time) to characterize temporal dynamics and subject-specific augmentations.

Key Features:

• Three-Way Data Analysis: Handles three-dimensional data structures such as patient×gene×time or subject×voxel×time and accommodates asynchronous longitudinal measurements.
• Core and Patient-/Subject-Specific Modules: Infers core modules shared across subjects and patient- or subject-specific augmentations that add elements (e.g., genes) to those core patterns.
• Hierarchical Bayesian Data Model: Represents the data using a hierarchical Bayesian framework to incorporate prior structure and quantify uncertainty.
• Gibbs Sampling Algorithm: Uses Gibbs sampling, a Markov Chain Monte Carlo technique, to explore posterior distributions and assign elements to modules.
• Performance and Validation: Demonstrated improved performance on simulated and real datasets, including gene expression time series from septic shock studies and resting-state fMRI identifying relevant brain regions.

Scientific Applications:

• Gene Expression Time-Series: Detects dynamic gene modules across patients and identifies patient-specific augmentations relevant to conditions such as septic shock.
• Functional MRI Analysis: Identifies voxel-time modules across subjects to locate brain regions associated with resting-state and task-related activity.

Methodology:

Inference is performed using a hierarchical Bayesian data model with Gibbs sampling (MCMC) to assign elements to core and subject-specific modules from three-way data.