| Gene discovery and characterization is a long and labor-intensive process. Gene co-expression network analysis is a long-standing powerful approach that can strongly enrich signals within gene expression datasets to predict genes critical for many cellular functions. Leveraging this approach with a large number of transcriptome datasets does not yield a concomitant increase in network granularity. Independently generated datasets that describe gene expression in various tissues, developmental stages, times of day, and environments can carry conflicting co-expression signals. The gene expression responses of the model C3 grass Brachypodium distachyon to abiotic stress is characterized by a co-expression-based analysis, identifying 22 modules of genes, annotated with putative DNA regulatory elements and functional terms. A great deal of co-expression elasticity is found among the genes characterized therein. An algorithm, dGCNA, designed to determine statistically significant changes in gene-gene co-expression relationships is presented. The algorithm is demonstrated on the very well-characterized circadian system of Arabidopsis thaliana, and identifies potential strong signals of molecular interactions between a specific transcription factor and putative target gene loci. Lastly, this network comparison approach based on edge-wise similarities is demonstrated on many pairwise comparisons of independent microarray datasets, to demonstrate the utility of fine-grained network comparison, rather than amassing as large a dataset as possible. This approach identifies a set of 182 gene loci which are differentially expressed under drought stress, change their co-expression strongly under loss of thermocycles or high-salinity stress, and are associated with cell-cycle and DNA replication functions. This set of genes provides excellent candidates for the generation of rhythmic growth under thermocycles in Brachypodium distachyon.. |
