Skip to main content

Table 7 Ablation study

From: Amogel: a multi-omics classification framework using associative graph neural networks with prior knowledge for biomarker identification

Experiment

BRCA

KIPAN

Acc. (STD)

F1 (STD)

AUROC (STD)

Acc. (STD)

F1 (STD)

AUROC (STD)

DNNa

75.58(2.7)

62.37(5.4)

90.05(1.0)

94.15(1.5)

94.12(2.2)

98.81(0.7)

FS(Dist1)_DNNb

81.56(3.1)

71.32(3.9)

95.86(1.0)

94.34(1.0)

93.02(1.9)

99.04(0.6)

FS(Dist2)_DNNc

77.66(2.1)

66.95(3.3)

90.51(1.5)

94.54(0.7)

93.45(2.9)

98.88(0.5)

ARM(Top1000)_DNNd

83.20(2.7)

71.78(4.9)

95.80(0.8)

94.95(2.2)

93.74(2.3)

99.03(0.3)

ARM(Top1000)_GNN(noPrior)e

83.03(1.2)

70.29(2.2)

94.26(1.3)

90.91(1.9)

90.92(2.0)

97.17(1.7)

ARM(Top1000)_GNN(allPrior)f

83.86(1.6)

74.94(1.3)

94.48(1.3)

95.49(1.4)

93.60(1.9)

97.18(1.3)

ARM(CBA)_GNN(allPrior)_DNNg

81.73(3.8)

72.86(9.5)

91.80(3.7)

95.35(0.7)

95.04(1.2)

99.05(0.4)

ARM(DNN)_GNN(allPrior)_DNNh

85.37(1.5)

74.39(5.5)

94.55(0.9)

95.35(1.2)

94.21(1.4)

99.14(0.4)

ARM(Top1000)_GNN(noPrior)_DNNi

83.29(3.5)

72.10(5.3)

95.46(1.6)

95.45(1.7)

94.32(0.9)

98.94(0.6)

ARM(Top1000)_GNN(allPrior)_DNNj

86.32(1.7)

75.67(4.4)

95.96(0.6)

96.06(1.4)

95.08(1.4)

99.37(0.4)

  1. Bold indicates the highest value and the bracket values are the standard deviation
  2. aConcatenated multi-omics datasets were passed into the DNN model for final classification
  3. bFeature selection using ANOVA-F method for each omics data independently
  4. The final distribution of features between omics is equally distributed. Total concatenated omics features are 450 for the BRCA dataset and 999 for the KIPAN dataset
  5. cMulti-omics data were concatenated, and subsequently, the features were reduced and selected using the ANOVA-F method. The final number of selected features is 450 for the BRCA dataset and 1000 for the KIPAN dataset
  6. dMulti-omics features were selected and concatenated using the ARM technique with Top-1000 ranked CARs. Selected features were passed into the DNN model for final classification
  7. eMulti-omics features were selected and concatenated from Top-1000 ranked CARs, generated using ARM technique. The synthetic information graph was constructed for final classification
  8. fMulti-omics features were selected and concatenated from Top-1000 ranked CARs, generated using ARM technique. Final static graphs were constructed from the information-based graph and prior knowledge graphs for graph feature learning. Graph feature learning and global feature learning were utilized for the final classification
  9. gMulti-omics features were selected and concatenated from pruned CARs by the CBA method. Final static graphs were constructed from the information-based graph and prior knowledge graphs for graph feature learning. Graph feature learning and global feature learning were utilized for the final classification
  10. hMulti-omics features were selected and concatenated from the best Top-K CARs based on DNN classifier performance. Final static graphs were constructed from the information-based graph and prior knowledge graph for graph feature learning. Graph feature learning and global feature learning were utilized for final classification
  11. iSingle dimensional edge graph for AMOGEL model, without prior knowledge information
  12. jThe proposed AMOGEL method
Back to article page

BMC Bioinformatics

ISSN: 1471-2105

Contact us