Email updates

Keep up to date with the latest news and content from BMC Bioinformatics and BioMed Central.

Open Access Highly Accessed Research article

GSA-PCA: gene set generation by principal component analysis of the Laplacian matrix of a metabolic network

Dan Jacobson* and Guy Emerton

Author Affiliations

Institute for Wine Biotechnology, Stellenbosch University, Stellenbosch, 7600, South Africa

For all author emails, please log on.

BMC Bioinformatics 2012, 13:197  doi:10.1186/1471-2105-13-197

Published: 9 August 2012

Abstract

Background

Gene Set Analysis (GSA) has proven to be a useful approach to microarray analysis. However, most of the method development for GSA has focused on the statistical tests to be used rather than on the generation of sets that will be tested. Existing methods of set generation are often overly simplistic. The creation of sets from individual pathways (in isolation) is a poor reflection of the complexity of the underlying metabolic network. We have developed a novel approach to set generation via the use of Principal Component Analysis of the Laplacian matrix of a metabolic network. We have analysed a relatively simple data set to show the difference in results between our method and the current state-of-the-art pathway-based sets.

Results

The sets generated with this method are semi-exhaustive and capture much of the topological complexity of the metabolic network. The semi-exhaustive nature of this method has also allowed us to design a hypergeometric enrichment test to determine which genes are likely responsible for set significance. We show that our method finds significant aspects of biology that would be missed (i.e. false negatives) and addresses the false positive rates found with the use of simple pathway-based sets.

Conclusions

The set generation step for GSA is often neglected but is a crucial part of the analysis as it defines the full context for the analysis. As such, set generation methods should be robust and yield as complete a representation of the extant biological knowledge as possible. The method reported here achieves this goal and is demonstrably superior to previous set analysis methods.