Inferring Genome-Scale Transcriptional Networks from Large-Scale Gene Expression Data

The accurate re-construction and interpretation of genome-scale Gene Association Networks (GANs) from large-scale gene expression data is challenging but crucial to the understanding of gene function, interaction, and cellular behavior at the genome level. Yet, reverse-engineering of genome-scale GANs from large-scale gene expression data and subsequently functional module discovery remains a very computationally intensive task; it generally requires both effective statistical approaches and efficient algorithm implementations through parallel computing engineering. Thus to date, it stands as one of the most pressing open problems in computational systems biology.

To facilitate genome-scale GAN reconstruction, we developed two new web servers: 1) GPLEXUS (http://plantgrn.noble.org/GPLEXUS/) and 2) DeGNServer (http://plantgrn.noble.org/DeGNServer/), both integrate a series of novel algorithms in a parallel-computing environment to construct and analyze genome-wide GANs from very large-scale expression profile datasets.

The GPLEXUS was specially designed for decipher genome-scale networks from large number of expression profiles for the organisms with medium to large size of genomes, such as those of most plant species. GPLEXUS adopts an ultra-fast estimation for pairwise mutual information computing which is similar in accuracy and sensitivity to the Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE) method and runs approximately one thousand times faster. GPLEXUS integrates Markov Clustering Algorithm (MCL) to effectively identify functional sub-networks. Furthermore, GPLEXUS includes a novel ‘condition-removing’ method to identify the major experimental conditions in which each sub-network operates.

The DeGNServer extended the widely used Context Likelihood of Relatedness (CLR) framework by the integration of six proven correlation/association analysis methods, e.g. Spearman rank correlation, Pearson correlation, Mutual-information, Maximum information coefficient, Kendall rank correlation, Thei-sen Estimator, and then implemented the algorithms in parallel and distributed environment to enable genome-scale gene-gene association inference from small to medium-sale gene expression profiles at extraordinary speed while maintaining high prediction accuracy. Furthermore, we integrated the SNBuilder and GeNa algorithms for sub-network extraction and functional module discovery.