We demonstrate the potential use of machine learning approaches in elucidating new insights into a biological response. Supervised machine learning methods are methods that group expression profiles into one of several known categories through comparison with known examples. Examples of supervised machine learning include logistic regression, neural networks, decision trees and linear discriminant analysis. The method we employ here is a version of linear discriminant analysis modified to cope with high-dimensional data. We used the publicly available NCBI60 cancer cell line microarray expression database as reported by Schref and colleagues1; this diverse data set includes samples from nine different tissue types. These cell lines were separated into high and low expressers of the Von Hippel Lindau gene (VHL). The VHL protein product is a critical regulator of the hypoxic response; it targets the hypoxia inducible factor 1-alpha (HIF-1a) transcription factor for proteasome-mediated degradation. We examined the features that the supervised machine learning algorithm used for classification and conjectured that they might be genes that are regulated at the level of transcription by HIF-1a. Many of these genes were already known HIF-1a target genes; others were genes known to be expressed under hypoxic conditions. Supervised machine learning methods succeed here where traditional unsupervised clustering methods may fail. The phenomenon studied here is restricted to a handful of genes, whose expression patterns may be masked by the tissue-specific profiles that will dominate in clustering.