As alternatives to experimental approaches, several groups have developed computational techniques for determining the targets of enhancers [7,16,21,66-70]. These methods are similar in that they compare patterns of regulatory activity across multiple cell types to predict interactions between enhancers and genes. However, they vary significantly in the type of data required to generate enhancer-gene predictions, the specific approaches used to generate and validate the predictions, and their availability (Table 1). The method described by Ernst and colleagues identifies H3K4me1/2 and H3K27ac sites that co-vary with expressed genes within 125 kb of the gene locus, and uses this to predict enhancer-gene interactions [21]. Thurman and co-workers utilized DHS exclusively to predict interactions. Enhancers were assigned to genes by correlating the cross-cell-type DNase I signal at each DHS site with all promoters located within 500 kb [66]. The method developed by Sheffield and colleagues also uses DHS profiles, but additionally incorporates genome-wide expression data [70]. Rather than employing a fixed distance-based model, Shen and colleagues apply chromatin conformation data from Hi-C experiments to guide the association of enhancers to genes marked by
