Skip to main content

Table 1 Network outputs for 25 blind test samples

From: Discovering biomarkers from gene expression data for predicting cancer subgroups using neural networks and relational fuzzy clustering

Sample # EWS RMS BL NB Actual
1 0.999 0.001 0.000 0.000 Sk. Muscle
2 0.000 0.224 0.000 0.004 Prostate Cancer
3 1.000 0.204 0.000 0.000 Sarcoma
4 0.000 0.000 0.000 1.000 NB
5 0.151 0.157 0.000 0.000 RMS
6 0.039 0.000 0.992 0.000 Sk. Muscle
7 0.016 0.000 0.999 0.000 Osteosarcoma
8 0.000 0.000 0.000 1.000 NB
9 1.000 0.001 0.000 0.008 EWS
10 0.000 1.000 0.000 0.000 RMS
11 0.003 0.000 1.000 0.000 BL
12 1.000 0.000 0.000 0.000 EWS
13 0.000 0.997 0.000 0.000 RMS
14 1.000 0.000 0.000 0.000 EWS
15 1.000 0.001 0.000 0.000 EWS
16 0.309 0.028 0.001 0.000 EWS
17 0.000 1.000 0.000 0.000 RMS
18 0.001 0.000 1.000 0.000 BL
19 0.000 1.000 0.000 0.000 RMS
20 0.000 0.000 0.000 1.000 NB
21 0.000 0.000 0.000 1.000 NB
22 0.000 0.000 0.000 1.000 NB
23 0.000 0.000 0.000 1.000 NB
24 0.000 0.000 1.000 0.000 BL
25 1.000 0.000 0.000 0.000 EWS
  1. The predicted and the actual outputs of a typical run for the 25 blind test samples. These 25 also includes the 5 nonSRBCT cases as marked in column 6 (sample numbers : 1,2,3,6, and 7). All but sample No. 16 of the 25 SRBCT test samples, the support provided by the network for the correct class is almost 1 and for the other classes it is practically 0. For sample No. 16, the support for the correct class is 0.309 but it is about 12 times stronger than the next high output value. Hence, this is also a decision with high confidence. For sample No. 5, although the network classified it correctly, the support for RMS and EWS are 0.157 and 0.151 suggesting further investigation (weak support for the RMS and the next higher support is for EWS, but that is very close to that of RMS.). Due to nonavailability of the patient's identity, this investigation could not be pursued further.