Gene set analysis is described in the paper by Efron and Tibshirani (2006):
On testing the significance of sets of genes (ps file)
(pdf file)

  • It differs from a Gene Set Enrichment Analysis (Subramanian et al 2006) in its use of the "maxmean" statistic: this is the mean of the positive or negative part of gene scores in the gene set, whichever is larger in absolute value.

  • Efron and Tibshirani shows that this is often more powerful than the modified Kolmogorov-Smirnov statistic used in GSEA.

  • GSA also uses a somewhat different null distribution for estimation of false discovery rates: it does "restandardization" of the genes (rows), in addition of the permutation of columns (done in GSEA), unlike GSEA (currently).

  • GSA also can handle phenotypes other than two groups, such as multiple classes, survival times and quantitative outcomes.