Pathway Distiller - multisource biological pathway consolidation

Abstract

Background: One method to understand and evaluate an experiment that produces a large set of genes, such as a gene expression microarray analysis, is to identify overrepresentation or enrichment for biological pathways. Because pathways are able to functionally describe the set of genes, much effort has been made to collect curated biological pathways into publicly accessible databases. When combining disparate databases, highly related or redundant pathways exist, making their consolidation into pathway concepts essential. This will facilitate unbiased, comprehensive yet streamlined analysis of experiments that result in large gene sets.

Methods: After gene set enrichment finds representative pathways for large gene sets, pathways are consolidated into representative pathway concepts. Three complementary, but different methods of pathway consolidation are explored. Enrichment Consolidation combines the set of the pathways enriched for the signature gene list through iterative combining of enriched pathways with other pathways with similar signature gene sets; Weighted Consolidation utilizes a Protein-Protein Interaction network based gene-weighting approach that finds clusters of both enriched and non-enriched pathways limited to the experiments' resultant gene list; and finally the de novo Consolidation method uses several measurements of pathway similarity, that finds static pathway clusters independent of any given experiment.

Results: We demonstrate that the three consolidation methods provide unified yet different functional insights of a resultant gene set derived from a genome-wide profiling experiment. Results from the methods are presented, demonstrating their applications in biological studies and comparing with a pathway web-based framework that also combines several pathway databases. Additionally a web-based consolidation framework that encompasses all three methods discussed in this paper, Pathway Distiller (http://cbbiweb.uthscsa.edu/PathwayDistiller), is established to allow researchers access to the methods and example microarray data described in this manuscript, and the ability to analyze their own gene list by using our unique consolidation methods.

Conclusions: By combining several pathway systems, implementing different, but complementary pathway consolidation methods, and providing a user-friendly web-accessible tool, we have enabled users the ability to extract functional explanations of their genome wide experiments.

Figure 1

Hierarchical clustering of expression data at gene level and pathway level. Hierarchical clustering of (A) absolute gene expression over 5 time points; and (B) pathway expression. Gene expression of samples at 5 different time points (TP1 to TP5) are marked above the heatmap, log2-transformed and normalized gene expression data of each gene were plotted in Figure 1A. Average of absolute gene expression over each pathway were evaluated and then plotted in Figure 1B to reflect the activities of biological pathways.

Figure 2

Hierarchical clusters of aggregated pathways. Dendrogram generated by using PPI similarity measurement and 250 clusters were colored automatically to illustrated grouping effect.

Figure 3

Pathway distiller screen shot of sample analysis. Steps for a typical resultant gene list analysis and pathway cluster discovery. Step 1: click to use sample resultant gene set. Step 2: click to use sample background set. Step 3: check all pathways. Step 4: click to do one-sided Fisher's Exact pathway enrichment; fill Enrichment Results grid. Step 5: click to do pathway consolidation; fills Consolidation Results grid; shows default consolidation method. Step 6: click to select from consolidation methods.

Figure 4

Interaction network for FOXA, FOXA1 and FOXA2/3 transcription factor networks. FOXA1 and AR interaction is noted (present/missing) to highlight if AR is in pathway or not.