Motivation: direct bearing whose expres Sorlie et al., 2 gene sets is a diseases (Lossos et al., 2004; Miklos and Maleszka, 2004), and the variables that could account questions in reasons such analysis. Results: To a t, and analyzed 't Veer et al., 2 fact the result t of patients used for gene selection. Many equally predictive lists could have been produced from e analysis. Three main properties of the data explain this sensitivity: (a orrelated with survival; (b) the differences between these correlati over different properties is d INTRODUC Several attem Tibshirani, 20 Rosenwald et 2003; Sorlie profiling. Sor clustering, to assign breast carcinoma tissues to one of five different subtypes, each with a distinctive expression profile. Robustness of these survival related subclasses was demonstrated(Sorlie et al., 2003) by applying the same analysis procedure to two independent breast carcinoma data sets(van 't Veer et al., 2002; West et al., 2001). van't Veer et al.(v...