Table 2.
Cross-validation experiments
| Factor | Criteria | CSI-Tree | ANOVA(0) | ANOVA(+) | MatrixREDUCE |
|---|---|---|---|---|---|
| Nkx-2.5 | MSE | 0.311 | 0.403 | 0.383 | 0.583 |
 |
0.709 | 0.509 | 0.549 | 0.368 | |
| PA1 | MSE | 1.662 | – | – | 3.405 |
 |
0.735 | – | – | 0.451 | |
| PA2 | MSE | 1.240 | 1.544 | 1.322 | 2.129 |
 |
0.740 | 0.740 | 0.764 | 0.685 |
ANOVA(0): ANOVA model with only main effect terms (each position of the binding site is contributing independently); ANOVA(+): ANOVA model where the model complexity, i.e. inclusion of higher order interactions, is based on a sequential F-test. MSE refers to averaged mean squared prediction error over the validation sets; 
refers to averaged Pearson correlation between the observed binding intensities and the predicted intensities over the validation sets. For PA1, cross-validation criteria for the ANOVA methods are not reported as ANOVA method does not provide prediction if a position in a sequence has a level (nucleotide) that has not been encountered in the training dataset.