A novel community detection method improves detection of functional gene modules in big gene expression data.

We identify communities of functionally related genes in the network inferred from the gene expression data of eukaryotic model organisms Arabidopsis thaliana & Saccharomyces cerevisiae by finding the network partition that maximizes the recently introduced generalized modularity density metric Q_g. This new metric does not suffer from the resolution limit problem and, with its tunable control parameter, can be used to study the hierarchical structure of communities. We use the Reduced Network Extremal Ensemble Learning (RenEEL) scheme [Sci. Rep. 9, 14234 (2019)] to optimize the metric. Statistical significance comparisons with the gene ontology indicate that the Q_g method outperforms other clustering methods. Orphan genes have been found in all sequenced species. These are genes unique to particular species. They are thought to play a key role in speciation, but their regulatory interactions remain largely unknown. Focusing on highly significant functional modules that contain orphan genes, regulatory interaction patterns involving these genes are discovered and testable predictions are made about their specific biological functions.