values less than 0.01 comprised the final hallmark set. When the number of genes obtained by this method was less than 15 (or more than 200) the top scoring 15 (or 200) genes were chosen regardless of their FDR values. Thus the refined hallmarks consist of at least 15 and at most 200 genes, which is the recommended size for use with GSEA. In the refinement procedure we focused on up-regulated genes and used one-tailed tests. The rationale for this stems from our empirical observation that expression patterns of down-regulated genes are often context dependent and tend to generalize poorly across datasets, while up-regulated genes are more consistent.
