Gene-miR pairs with negative expression correlations are overrepresented in predicted target databases. Pearson correlations were calculated for the subset of gene-miR pairs found in the miRanda and TargetScan databases, and binned by correlation value. For each bin, the ratio of observed to expected pairs (odds ratio) was calculated. Bins which have a positive log odds ratio are over-represented in predicted targeting pairs, and bins with negative log odds ratio are under-represented. miR-gene pairs with negative expression correlation are clearly enriched in miRanda predicted interacting pairs, whereas TargetScan shows a bimodal enrichment pattern. A) miRanda conserved, B) miRanda nonconserved, C) TargetScan.