Statistical aspects of gene signatures and molecular targets.

Evolution of high-throughput technologies has enabled us to quantify several thousands of gene expressions simultaneously. Several oncologic applications have emerged, two of which will be discussed: developing gene signatures and finding molecular targets. This article emphasizes the different nature of statisitical methods used and required for developing gene signatures and identifying molecular targets. Routine and careful consideration of validation methods are essential for developing gene signatures. Choice of model development methodology seems to be less critical. Because of the way gene signatures are developed, even after careful validation, it is unlikely that they will yield a direct link to molecular targets. Identifying molecular targets will require more careful and focused experimentation than large-sample microarray studies.