Skip to main content

Table 2 Computational methods for target gene prediction

From: A curated benchmark of enhancer-gene interactions for evaluating enhancer-target gene prediction methods

Unsupervised methods
 DistanceRanks pairs by inverse linear distance 
 DNase-DNaseCalculates the Pearson correlation coefficient between the DNase signals at enhancers and promoters across 32 cell-type categories.[22]
 DNase-expressionCalculates the Pearson correlation coefficient between the normalized DNase signals at enhancers and normalized gene expression levels measured by microarray across 112 cell types.[23]
 GeneHancerCell-type agnostic predictions based on co-expression correlations, CHi-C interactions, eQTLs, and genomic distance[31]
 Average-rankCombines the distance and DNase-expression methods by averaging the rank of for each prediction between the two methods 
Supervised methods
 PEP-motifFeatures: frequency of motif instances at enhancers and promoters[28]
Classifier: Gradient boosting (XGB package)
 TargetFinderFeatures: Cell-type-specific epigenomic signals (ChIP-seq, DNase, CAGE, etc.) at enhancers, promoters, and the intervening window between enhancers and promoters.[27]
Classifier: Gradient boosting (scikit learn)