Pattern-Based Gene-Set Recognition for Interpreting Genome-Wide Gene Expression Profiles
Abstract
Background:
Accurate recognition of important gene sets from genome-wide gene expression profiles provides great insights into the underlying biological mechanisms that govern the gene expression dynamics. However, most gene set recognition algorithms rely solely on supervised sample phenotypic information, overlooking the unsupervised gene-gene expression correlations that are inherently informative in the gene expression profiles.
Results:
We developed a computational framework named PAGER (Pattern Acquisition and GEne-set Recognition) for identifying gene sets showing significant supervised and unsupervised patterns. We showcased the use of PAGER in several recent expression profiling studies including cadmium treated rat primary hepatocyte toxicogenomics study and adrenal gland periodical gene expression profiling. Our results indicate that PAGER achieved better performance in discovering truly important pathways from expression profiles which were undetected using current other existing tools. These results were further corroborated by literature and cytotoxicity experiments.
Conclusions:
PAGER integrated both supervised and unsupervised pattern metrics for gene set summarization. For each given gene set, PAGER provides a two-dimensional view showing its external activity and internal coherence pattern. PAGER employed statistical methods such as Relaxed Intersection-Union Tests, Stouffer’s method and Fisher’s method for integration of pattern significance. In addition, PAGER can be used for recognizing user-defined arbitrary gene set as demonstrated in one of our previous publications. PAGER is freely available for academic user at http://dengx.bol.ucla.edu/PAGER/PAGER.htm.