Skip to main content

Table 5 Characterization of each method's performances

From: Functional Analysis: Evaluation of Response Intensities - Tailoring ANOVA for Lists of Expression Subsets

  a2.fixed faeri.
fixed.null
faeri.
fixed.perms
GSA.
mean*
GSA.
absmean*
GSA.
maxmean*
Globaltest.
asymptotic
Globaltest.
gamma
Globaltest.
perm
Gsea.
pval
Gsea.
fdr
Samgs.
pval
100% uni 100% 99% 99% 57% 62% 56% 97% 99% 98% 99% 88% 99%
50% uni 94% 87% 89% 50% 51% 50% 71% 9'% 92% 71% 62% 90%
100%uni + cor 94% 84% 85% 50% 51% 51% 60% 67% 62% 58% 51% 61%
50%uni + cor 82% 76% 77% 50% 50% 50% 59% 65% 62% 56% 57% 61%
100% bidir 50% 100% 100% 50% 83% 70% 100% 100% 100% 55% 50% 100%
50% bidir 50% 97% 97% 50% 53% 51% 96% 100% 100% 50% 50% 100%
100% bidir + cor 50% 94% 94% 50% 55% 54% 69% 75% 71% 58% 54% 71%
50% bidir + cor 50% 87% 88% 50% 51% 50% 68% 75% 71% 51% 50% 70%
Mean 71% 90% 91% 51% 57% 54% 77% 84% 82% 62% 58% 81%
Rank 7 2 1 12 10 11 6 3 4 8 9 5
  1. Comparison of the number of true positives and true negatives to the total number of tests (VP+VN)/(P+N). For information, we also computed the mean value for the 8 types of groups studied. The significance threshold to select groups was set at 0.01. The highlighted (in bold) values are relative to methods obtaining a score of more than 75%. The four global methods are more appropriate for geneset analysis and the two-step methods show poor performance. Among the global methods, only ANOVA-2 and FAERI happen to be suited for unidirectional correlated genesets. FAERI and SAMGS are the only methods adapted for the study of bidirectional correlated genesets. Across all genesets tested, the methods giving the best results were: FAERI (perms), FAERI (null), SAMGS (q-value), GlobaTest (gamma), GlobaTest (perms), SAMGS (p-value), GlobalTest (asymptotic). FAERI is the only method suited for the study of all types of genesets tested.