Bioinformatics Advance Access published online on August 20, 2008
Bioinformatics, doi:10.1093/bioinformatics/btn440
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Local Coherence in Genetic Interaction Patterns Reveals Prevalent Functional Versatility
1Molecular structure and Function Program, Hospital for Sick Children, 555 University Avenue, Toronto ON, Canada M5G 1X8
2Terrence Donnelly Centre for Cellular & Biomolecular Research, 160 College Street, Toronto ON, Canada, ON M5S 3E1
3Department of Molecular Genetics, University of Toronto, 1 King's College Circle, Toronto ON, Canada M5S 1A8
4Department of Biochemistry, University of Toronto, 1 King's College Circle, Toronto ON, Canada M5S 1A8
*To whom correspondence should be addressed. Dr. Shuye Pu, E-mail: shuyepu{at}sickkids.ca
| Abstract |
|---|
Motivation: Epistatic or genetic interactions, representing the effects of mutating one gene on the phenotypes caused by mutations in one or more distinct genes, can be very helpful for uncovering functional relationships between genes. Recently, the Epistatic Miniarray Profiles (E-MAP) method has emerged as a powerful approach for identifying such interactions systematically. For E-MAP data analysis, hierarchical clustering is used to partition genes into groups on the basis of the similarity between their global interaction profiles, and the resulting descriptions assign each gene to only one group, thereby ignoring the multi-functional roles played by most genes.
Results: Here we present an original Local Coherence Detection (LCD) algorithm for identifying groups of functionally related genes from E-MAP data in a manner that allows individual genes to be assigned to more than one functional group. This enables investigation of the pleiotropic nature of gene function. The performance of our algorithm is illustrated by applying it to two E-MAP datasets and an E-MAP-like in silico dataset for the yeast S. cerevisiae. In addition to recapitulating the majority of the functional modules and many protein complexes reported previously, our algorithm uncovers many recently documented and novel multi-functional relationships between genes and gene groups. Our algorithm hence represents a valuable tool for uncovering new roles for genes with annotated functions and for mapping groups of genes and proteins into pathways.
Availability: A Java implementation of the LCD algorithm is available at URL http://genepro.ccb.sickkids.ca/biclustering.html.
Contact: shuyepu{at}sickkids.ca
Supplementary information: URL to be provided by the journal
Associate Editor: Dr. Trey Ideker
Received on May 13, 2008; revised on August 1, 2008; accepted on August 15, 2008