Table 2. Effect of cooperative interactions between pairs of TFs on the accuracy of modeling ChIP data.
| Data set | M2 | CC(M1) | CC(M2) | CC(M1+M2) | ImprOverM1 | ImprOverM2 | P-value | Z-score |
| BCD_Bseq_s5 | VFL | 0.560 | 0.392 | 0.603 | 0.043 | 0.211 | 0.00 | 9.03 |
| BIN_Fchip_s10_11 | TRL | 0.630 | 0.206 | 0.682 | 0.052 | 0.475 | 0.02 | 3.91 |
| BIN_Fchip_s14 | TRL | 0.654 | 0.216 | 0.697 | 0.043 | 0.481 | 0.01 | 7.64 |
| CAD_Bseq_s5 | VFL | 0.178 | 0.472 | 0.519 | 0.341 | 0.048 | 0.00 | 9.99 |
| D_Bchip_s5 | VFL | 0.291 | 0.310 | 0.393 | 0.102 | 0.083 | 0.00 | 19.97 |
| D_Mseq_s5_11 | HLHM5 | 0.328 | 0.373 | 0.416 | 0.088 | 0.043 | 0.04 | 4.00 |
| DA_Bchip_s5 | VFL | 0.286 | 0.266 | 0.334 | 0.048 | 0.068 | 0.00 | 11.84 |
| EN_Mchip_s5_14 | TTK | 0.321 | −0.159 | 0.391 | 0.070 | 0.232 | 0.01 | 23.81 |
| HB_Bseq_s5 | VFL | 0.374 | 0.326 | 0.459 | 0.085 | 0.133 | 0.00 | 6.09 |
| KR_Bseq_s5 | VFL | 0.430 | 0.388 | 0.531 | 0.101 | 0.143 | 0.00 | 21.21 |
| MED_Bchip_s14 | ZIF | 0.255 | 0.105 | 0.337 | 0.082 | 0.232 | 0.01 | 3.33 |
| RUN_Bchip_s5 | TRL | 0.258 | 0.147 | 0.328 | 0.070 | 0.181 | 0.00 | 12.72 |
| SLP1_Bchip_s5 | VFL | 0.214 | 0.154 | 0.281 | 0.067 | 0.128 | 0.00 | 3.47 |
| TIN_Fchip_s10_11 | TRL | 0.392 | 0.269 | 0.472 | 0.080 | 0.203 | 0.00 | 29.57 |
| TIN_Fchip_s9 | TRL | 0.428 | 0.257 | 0.480 | 0.052 | 0.223 | 0.02 | 15.11 |
| TWI_Bchip_s5 | VFL | 0.251 | 0.280 | 0.345 | 0.094 | 0.065 | 0.00 | 15.84 |
| TWI_Fchip_s10_11 | TRL | 0.405 | 0.305 | 0.465 | 0.060 | 0.160 | 0.02 | 10.41 |
| TWI_Fchip_s9 | TRL | 0.420 | 0.197 | 0.469 | 0.049 | 0.272 | 0.00 | 10.55 |
STAP was used with two motifs – the primary motif (representing the ChIP'ed TF) and a secondary motif (“M2”), and with cooperative DNA-binding included in the model. Cooperative interaction between two TF sites was included in the model only if the two sites are within a pre-defined “Distance Threshold” set to 150 bp in this set of experiments. The correlation coefficient between STAP scores from this model (CC(M1+M2)) was compared to the CC when using only the primary motif (CC(M1)) or when using only the secondary motif (CC(M2)) in STAP. The respective improvements are noted as “ImprOverM1” and “ImprOverM2” respectively. The column “P-value” shows an empirically calculated p-value for the improvement, comparing the observed improvements to that expected from 100 shuffled versions of the secondary motif. The column “ImprOverM2” is the difference of CC(M1+M2) and the absolute value of CC(M2). The last column (“Z-score”) compares the observed improvement (ImprOverM1) to that obtained using other real motifs, corresponding to TFs expressed highly in that developmental stage, as the secondary motif. Shown here is only the single strongest secondary motif influence on each data set, if its P-value is ≤0.05 and Z-score is ≥3. The complete list of significant effects is in Supplementary Materials (Table S2). Note that all results are from cross-validation and thus account for the additional parameters in the two motifs model compared to the one motif model.