Skip to main content

Table 2 Balanced accuracies of prediction using the metagene classifier (ROC) and various classification algorithms and a t-test as feature selection criteria.

From: Robust assignment of cancer subtypes from expression data using a uni-variate gene expression average as classifier

Phenotype/
Method
Training1 Validation Validation Training Validation Validation Training Validation Validation Mean
ER 2 (BC) vandeVijver WangY Sotiriou WangY vandeVijver Sotiriou Sotiriou vandeVijver WangY  
ROC5 0.95 0.89 0.74 0.89 0.94 0.77 0.77 0.90 0.88 0.85
SVM 0.94 0.82 0.71 0.82 0.95 0.73 0.69 0.95 0.82 0.83
SVMradial 0.94 0.86 0.70 0.87 0.93 0.73 0.76 0.93 0.84 0.83
SVMpoly 0.92 0.80 0.67 0.77 0.89 0.65 0.75 0.84 0.72 0.76
knn 0.94 0.83 0.70 0.84 0.93 0.78 0.77 0.95 0.78 0.83
rforest 0.98 0.85 0.73 0.84 0.93 0.73 0.76 0.98 0.83 0.84
rpart 1.00 0.79 0.61 0.82 0.90 0.63 0.72 1.00 0.75 0.78
bagging 1.00 0.84 0.72 0.86 0.91 0.72 0.81 1.00 0.84 0.84
lda 0.95 0.85 0.76 0.86 0.94 0.73 0.76 0.96 0.83 0.84
dlda 0.96 0.89 0.79 0.88 0.94 0.77 0.87 0.94 0.85 0.86
slda 0.94 0.86 0.74 0.85 0.94 0.77 0.77 0.94 0.85 0.85
neuralnet 0.94 0.78 0.69 0.81 0.92 0.72 0.76 0.97 0.80 0.81
ncc 0.96 0.89 0.78 0.88 0.94 0.75 0.82 0.95 0.87 0.86
MYCN 3 (NB) WangQ JanoueixL Attiyeh JanoueixL WangQ Attiyeh Attiyeh WangQ JanoueixL  
ROC 0.94 0.91 0.82 0.87 0.89 0.79 0.81 0.95 0.89 0.88
SVM 0.98 0.83 0.75 0.88 0.97 0.67 0.85 0.98 0.87 0.84
SVMradial 0.95 0.93 0.82 0.93 0.94 0.81 0.87 0.95 0.86 0.88
SVMpoly 0.95 0.86 0.74 0.79 0.88 0.69 0.82 0.90 0.79 0.81
knn 0.95 0.93 0.86 0.88 0.92 0.85 0.86 0.98 0.85 0.90
rforest 0.95 0.93 0.80 0.91 0.94 0.76 0.84 0.95 0.88 0.88
rpart 1.00 0.93 0.90 0.83 0.89 0.72 0.90 1.00 0.72 0.86
bagging 1.00 0.90 0.89 0.81 0.91 0.72 0.87 1.00 0.86 0.88
lda 1.00 0.70 0.83 0.69 0.94 0.73 0.86 0.98 0.65 0.80
dlda 0.94 0.90 0.86 0.90 0.94 0.86 0.84 0.95 0.85 0.89
slda 0.95 0.93 0.85 0.93 0.98 0.83 0.89 0.98 0.91 0.91
neuralnet 0.99 0.81 0.89 0.85 0.96 0.80 0.88 0.97 0.82 0.87
ncc 0.94 0.91 0.86 0.93 0.94 0.85 0.86 0.98 0.87 0.90
AD/SQ 4 (LC) Angulo Takeuchi   Takeuchi Angulo      
ROC 0.93 0.96   0.95 0.96      0.96
SVM 0.89 0.91   0.93 0.84      0.88
SVMradial 0.97 0.93   0.96 0.96      0.95
SVMpoly 0.89 0.92   0.90 0.96      0.94
knn 0.94 0.94   0.95 0.94      0.94
rforest 0.94 0.95   0.96 0.94      0.94
rpart 0.86 0.93   0.95 0.82      0.87
bagging 0.91 0.96   0.95 0.93      0.94
lda 0.80 0.91   0.93 0.83      0.87
dlda 0.94 0.95   0.95 0.92      0.93
slda 0.93 0.95   0.95 0.96      0.96
neuralnet 0.91 0.95   0.94 0.92      0.94
ncc 0.93 0.96   0.95 0.93      0.95
  1. 1 Accuracies given in italics indicates accuracies obtained in the training set using LOOCV
  2. 2 ER+ vs. ER-
  3. 3 MYCN amplification positive vs. MYCN amplification negative
  4. 4 adenocarcinoma vs. squamous cell carcinoma
  5. 5 the metagene classifier uses AUC values as feature selection criteria.
  6. Abbreviations; BC = Breast Cancer, NB = Neuroblastoma, LC = Lung Cancer