fastBMA

fastBMA infers gene regulatory networks from genome-wide expression data using Bayesian Model Averaging to enable efficient and accurate reconstruction of genetic interactions.

Key Features:

• Bayesian Model Averaging: Implements an advanced Bayesian Model Averaging (BMA) framework for network inference from expression data.
• Distributed, Parallel, and Scalable Architecture: Executes in parallel across multiple CPU cores and supports distributed/cloud deployment to handle large genome-wide datasets.
• Transitive Reduction Module: Eliminates redundant indirect edges by mapping the transitive reduction problem to a shortest-path problem for computational efficiency.
• Performance and Efficiency: Demonstrates superior speed and reduced memory usage, with reported performance up to 100 times faster than LASSO on a single CPU core and validated on synthetic and genome-wide time series datasets from yeast and humans.
• Scalability: Capable of constructing large networks, exemplified by building a 10,000-gene regulation network in a few hours on a 32-core cloud cluster (two nodes of 16 cores each).
• Improved Accuracy and Speed: Provides enhanced accuracy and faster execution compared to ScanBMA and other rapid network inference methods.

Scientific Applications:

• Systems biology: Reconstruction of gene regulatory networks to study regulatory mechanisms and interactions in complex biological systems.
• Genome-wide time-series analysis: Inference of regulatory relationships from time-series expression datasets, including applications in yeast and human genomes.

Methodology:

Computational methods explicitly include Bayesian Model Averaging for model inference, parallel and distributed execution across CPU cores, and a transitive reduction step implemented by mapping to a shortest-path problem.