Functional genomics experiments can only have biological or clinical relevance if information (or better yet, knowledge) external to the system being studied is employed. Even in the case of so-called unsupervised algorithms or clustering techniques, the early investigations (e.g., [56, 63]) made sense of the clusters of expression only by reference to external knowledge: the genes known to be involved in cell cycle regulation, protein translation, etc. Without such external a priori knowledge, the observed clusters are of little use in generating or testing new hypotheses about biology. In the experiments that had the most obvious clinical relevance, such as finding genes that aid in the distinction between acute myelogenous leukemia versus acute lymphoblastic leukemia , the prognosis of large B-cell lymphoma , or pharmacogenomic prediction of gene targets for chemotherapy , it was central to the import of these investigations that they integrated well-characterized external biological or clinical data with the expression data.
If functional genomics is to lead to a qualitative change in the way in which clinical medicine (diagnostics, prognostics, and therapeutics) is practiced, then most studies will similarly have to incorporate high-quality and relevant external data. That is, if we want to know how a gene expression pattern may predict mortality, a particular disease profile, or drug responsiveness, we need to be at least as meticulous in characterizing the patients, their histories, their tissues, and the contexts surrounding how data were acquired, as we are in obtaining the gene expression profile. Perhaps because the need has only become recently apparent, to date there has been little in the way of systematic approaches to the acquisition of such extragenomic data. We discuss here why this is and what steps can be taken toward a more systematic approach.
Was this article helpful?