Skip to main content

Table 3 Subtype prediction accuracy of 5 gene lists in the GB, BC and OC validation datasets

From: Splitting random forest (SRF) for determining compact sets of genes that distinguish between cancer subtypes

Cancer Methods Classical (N = 50) Mesenchymal (N = 48) Neural (N = 30) Proneural (N = 48)
GB SRF50 70.0% 81.3% 73.3% 93.8%
  Single RF 64.0% 77.1% 80.0% 91.7%
  Verhaak et al. 88.0% 91.7% 50.0% 100.0%
  ANOVA 88.0% 91.7% 40.0% 100.0%
  Top 50 ANOVA 66.0% 83.3% 73.3% 87.5%
   Basal-Like (N = 18) HER2 (N = 11) Luminal A (N = 20) Luminal B (N = 32)
BC SRF50 100.0% 63.6% 80.0% 84.4%
  Single RF 88.9% 72.7% 55.0% 87.5%
  Parker et al. 100.0% 54.5% 90.0% 93.8%
  ANOVA 100.0% 36.4% 85.0% 93.8%
  Top 50 ANOVA 100.0% 63.6% 75.0% 84.4%
   C1 (N = 42) C2 (N = 25) C4 (N = 23) C5 (N = 18)
OC SRF50 97.6% 88.0% 73.9% 94.4%
  Single RF 97.6% 88.0% 78.3% 83.3%
  Tothill et al. 97.6% 88.0% 87.0% 88.9%
  ANOVA 100.0% 88.0% 78.3% 88.9%
  Top 50 ANOVA 95.2% 88.0% 73.9% 83.3%