Parallel Mutual Information Based Construction of Genome-Scale Networks on the Intel® Xeon Phi™ Coprocessor

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Whole-genome network construction from gene expression data is crucial but computationally intensive.
• Existing methods often struggle with scalability or require extensive computing clusters.

Purpose of the Study:

• To present a scalable solution for whole-genome network reconstruction using parallel computing.
• To demonstrate efficient network inference on specialized hardware like Intel Xeon Phi.

Main Methods:

• Utilized TINGe, a parallel network reconstruction technique.
• Employed mutual information and permutation testing for statistical significance.
• Leveraged multi-level parallelism on Intel Xeon Phi and Xeon processors.

Main Results:

• Successfully inferred the first whole-genome regulatory network for *Arabidopsis thaliana* on a single chip.
• Constructed a 15,575-gene network from 3,137 microarray experiments in just 22 minutes.
• Demonstrated significant performance gains through parallel mutual information computation.

Conclusions:

• The developed method enables efficient, large-scale whole-genome network inference.
• Specialized hardware and parallel algorithms can overcome computational bottlenecks in systems biology.
• Optimization techniques are transferable to other computational domains.