fastBMA: scalable network inference and transitive reduction.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Inferring genetic regulatory networks from genome-wide expression data presents significant computational challenges.
• Existing methods often struggle with scalability and speed when analyzing large datasets.

Purpose of the Study:

• To introduce fastBMA, a novel, efficient, and scalable implementation of Bayesian model averaging (BMA) for genetic network inference.
• To enhance computational performance and accuracy in reconstructing gene regulatory networks.

Main Methods:

• Developed fastBMA, a distributed and parallel implementation of Bayesian model averaging (BMA).
• Integrated a computationally efficient module for transitive reduction to eliminate redundant network edges via a shortest-path algorithm.
• Evaluated performance on synthetic and experimental yeast and human genome-wide time series expression data.

Main Results:

• fastBMA demonstrates up to 100x speed improvement over LASSO on a single CPU core with increased accuracy.
• The method is memory-efficient and scales effectively to human genome-wide expression data.
• A 10,000-gene network can be inferred in hours using a 32-core cluster, significantly outperforming ScanBMA and LASSO.

Conclusions:

• fastBMA offers a substantial advancement in speed and accuracy for genetic network inference.
• Its scalability enables timely analysis of large-scale genomic data.
• The integrated transitive reduction improves accuracy, particularly in denser networks.