Skip to main content

Table 3 Classification performance within breast cancer datasets

From: Improving biomarker list stability by integration of biological knowledge in the learning process

  GSE2990 GSE3494 GSE7390
No prior 0.89 (95%)
7
0.93 (93%)
10
0.90 (98%)
6
GO BP 0.62 (93%)
15
0.63 (95%)
21
0.60 (96%)
13
GO MF 0.63 (93%)
17
0.68 (94%)
24
0.60 (97%)
15
PPI NG 0.57 (94%)
10
0.58 (96%)
14
0.53 (97%)
9
PPI JA 0.87 (95%)
7
0.91 (93%)
11
0.87 (97%)
7
PPI FS 0.88 (95%)
7
0.92 (95%)
11
0.88 (97%)
7
PPI SC 0.83 (95%)
8
0.86 (95%)
13
0.83 (96%)
8
PE 0.78 (95%)
49
0.89 (96%)
56
0.79 (96%)
37
SP 0.78 (95%)
48
0.89 (95%)
56
0.79 (95%)
38
MI 0.76 (91%)
130
0.80 (94%)
207
0.73 (94%)
131
  1. Normalized Canberra distance between feature lists obtained for datasets GSE2990, GSE3494 and GSE7390, using the standard classification approach without prior knowledge integration and different prior knowledge based similarity matrices: Gene Ontology Biological Process (GO BP), Gene Ontology Molecular Function (GO MF), protein-protein interactions codified by the normalized geodesic distance (PPI NG), the Jaccard coefficient (PPI JA), the functional similarity (PPI FS), the probabilistic common neighborhood similarity (PPI SC), the Pearson correlation (PE), the Spearman rank correlation (SP) and the Mutual Information (MI). Predictive accuracy is indicated in brackets, whereas the number of iterations obtained by the classifier is reported below the other scores.