Table 2. SPI values of different classification systems on the nonsmoker and KRAS datasets.
| Feature set (#DEGs/DETs/DEEs) | Non-smoker (24 samples) |
KRAS (16 samples) |
|---|---|---|
| DESeq DEGs (13) | 1.000 | 0.996 |
| Cuffdiff DEGs (14) | 1.000 | 0.991 |
| Cuffdiff DETs (50) | 0.995 | 0.948* |
| DEXseq DEEs (397) | 0.998 | 0.919* |
| DEXseq DEEs – Naïve FPKM (28) | 0.930 | 1.000 |
*
In the KRAS dataset, the DET- and DEE-based systems each made an error by assigning a tumor sample to the “normal” group, even though the SPI values were fairly high. All other predictions in this table reached 100% accuracy.