Comprehensive enhancer-target gene assignments improve gene set level interpretation of genome-wide regulatory data

Abstract

Revealing the gene targets of distal regulatory elements is challenging yet critical for interpreting regulome data. Experiment-derived enhancer-gene links are restricted to a small set of enhancers and/or cell types, while the accuracy of genome-wide approaches remains elusive due to the lack of a systematic evaluation. We combined multiple spatial and in silico approaches for defining enhancer locations and linking them to their target genes aggregated across >500 cell types, generating 1,860 human genome-wide distal Enhancer to Target gene Definitions (EnTDefs). To evaluate performance, we used gene set enrichment testing on 87 independent ENCODE ChIP-seq datasets of 34 transcription factors (TFs) and assessed concordance of results with known TF Gene Ontology (GO) annotations., assuming that greater concordance with TF-GO annotation signifies better enrichment results and thus more accurate enhancer-to-gene assignments. Notably, the top ranked 741 (40%) EnTDefs significantly outperformed the common, na&iumlve approach of linking distal regions to the nearest genes (FDR < 0.05), and the top 10 ranked EnTDefs performed well when applied to ChIP-seq data of other cell types. These general EnTDefs also showed comparable performance to EnTDefs generated using cell-type-specific data. Our findings illustrate the power of our approach to provide genome-wide interpretation regardless of cell type.

Regulatory genome annotation of 33 insect species

“…There are clear shortcomings to this approach, as many enhancer targets are not the closest genes, leading to mis-assignments (e.g. 24,[25][26][27]. However, potentially more accurate methods for target-gene assignment rely on gene expression data, epigenetic data, and/or chromatin conformation data that are frequently not available for the species we are studying and thus difficult to incorporate into our prediction pipeline (27)(28)(29)(30).…”

“…24,[25][26][27]. However, potentially more accurate methods for target-gene assignment rely on gene expression data, epigenetic data, and/or chromatin conformation data that are frequently not available for the species we are studying and thus difficult to incorporate into our prediction pipeline (27)(28)(29)(30). Putative target genes are then mapped to their D. melanogaster orthologs (if an ortholog exists) using the Orthologer software from the Zdobnov lab (31) (see Methods).…”