Table 2 Overlap between ChIP-on-chip events with P ≤ 0.005 (or P ≤ 0.005 excluding high confidence binding events with P ≤ 0.001 and with sequence conservation across at least 3 yeast species) and predicted links obtained by the LINK model and by other five reference models. The table shows that less than 1% of all possible randomly selected links occur in binding experiments. Moreover, only 3.32% of the TF-gene pairs with overall expression correlation of greater (or smaller) than 0.5 (or -0.5) overlap with binding experimental data. The overlap increases to 13.37% when we consider highly correlated TF-gene pairs under the experimental conditions of the UTMs generated by the LINK model. By filtering the predicted links via the PWM matching, we discriminate between direct and indirect predicted interactions of co-expressed TF-gene pairs. As shown in the table, 40% (14.37%) of the new links predicted by the LINK model overlap with ChIP-on-chip binding all events with P ≤ 0.005 (P ≤ 0.005 excluding high confidence binding events with P ≤ 0.001 and with sequence conservation across at least 3 yeast species). This is a substantially higher rate than the 23.19% (7.25%) rate obtained by using a simpler approach that combines correlation with PWM-promoter matching, but disregards information about other experimentally known links.

From: Unraveling condition specific gene transcriptional regulatory networks in Saccharomyces cerevisiae