Biology Reference
In-Depth Information
patients. Hartung
54
has shown that, by modifying the weights of the
inverse normal method, dependent statistics can be accommodated.
Thus, our procedure should readily extend to this more general case.
We have found the concept of coexpression modules to be a versatile
tool for biologically unifying disparate results and producing an inte-
grated gene signature. Although coexpression does not imply direct
physical interactions, the highly correlated genes in a module can be con-
sidered surrogate markers of one another and of the same underlying
transcriptional process. Thus, coexpression is more appropriate for
understanding the equivalence of signatures than for functional annota-
tions of the genes. Coexpression modules can also be used to dissect sig-
natures, revealing the parts that are essential. The strong correlation of
expression within a module allows summarizing the module's overall
expression by simple averaging. These module scores concisely and
robustly characterize a tumor by a handful of quantitative measures with
straightforward interpretation.
In summary, we provide a unified, flexible, and extensible framework
for integrated analysis of heterogeneous genomic datasets. We have
demonstrated here how this framework can be used to unify results of
previous gene expression studies in breast cancer. These methods are
practical and widely applicable to a variety of technologies and biological
investigations.
Acknowledgments
This work was supported by the European Commission Framework
Programme VI (FP6-LSHC-CT-2004-503426) and by the Swiss National
Science Foundation through the National Centres for Competence in
Research in Plant Survival and Molecular Oncology.
References
1. Edgar R, Domrachev M, Lash AE. (2002) Gene Expression Omnibus: NCBI
gene expression and hybridization array data repository.
Nucleic Acids Res
30
: 207-10.
