Linking enhancers to target genes

We have developed leading computational models to predict which enhancers regulate which genes in the human genome, and constructed initial maps across thousands of cell types.

Explore enhancer maps across 1000s of cell types here:

E2G Browser — e2g.stanford.edu
IGVF Catalog — catalog.igvf.org

Apply model to build enhancer maps in new cell types:

ENCODE-rE2G model – For cell types in which you have bulk ATAC-seq or DNase-seq, plus optionally H3K27ac and Hi-C
scE2G model – For single-cell ATAC-seq or multiomic single-cell RNA+ATAC-seq
ABC model – The Activity-by-Contact (ABC) model is now incorporated into ENCODE-rE2G and scE2G

Evaluate new models using genomic perturbation data:

CRISPR benchmarking pipeline — How well do models predict results of CRISPR perturbations to distal elements?
eQTL benchmarking pipeline — How well do models link fine-mapped eQTL variants to target eGenes?
GWAS benchmarking pipeline — How well do models link fine-mapped GWAS variants to high-confidence target genes?

ENCODE-rE2G overview: models are trained on enhancer–gene features and CRISPR-validated pairs, benchmarked against CRISPR screens, fine-mapped eQTLs and GWAS variants, then applied across 1,458 DNase-seq experiments to map enhancer–gene regulatory interactions.

Fetal Heart Map

The Fetal Heart Map is a single-cell multiomic atlas of the developing human heart: gene expression and chromatin accessibility in ~730,000 nuclei from 41 fetal hearts (6–22 post-conception weeks), annotated into more than 90 cardiac cell types and states. Browse which genes each cell type expresses, and view chromatin accessibility and enhancer–gene prediction tracks — the maps we used to interpret variants associated with congenital heart disease.

Explore the data:

Fetal Heart Map browser

Publications:

Ma, Conley, and Chen et al. medRxiv (2024). Interpreting variants associated with congenital heart disease with a regulatory map of the human fetal heart.

Fetal Heart Map: a UMAP of fetal heart nuclei colored by cell type (cardiomyocytes, fibroblasts, endothelium, endocardium, conduction, neural crest, immune and epicardial cells) beside a spatial map of a fetal heart section; below, a schematic linking variants to regulatory elements, genes, cell types and pathways in congenital heart disease (CHD).

Activity-by-Contact (ABC) Model

The Activity-by-Contact (ABC) Model predicts which enhancers regulate which genes in the genome, based on estimating enhancer activity and enhancer-promoter contact frequency from epigenomic datasets.

Publications:

Fulco and Nasser et al. Nature Genetics (2019). Initial description and validation of the ABC model. See Supplementary Methods for details of the implementation.
Nasser et al. Nature (2021): Genome-wide enhancer maps connect risk variants to disease genes. Produces ABC predictions across 131 biosamples and describes their utility in interpreting fine-mapped GWAS variants.

Data:

ABC predictions in 131 cell types and tissues — all element-gene connections with ABC scores ≥ 0.015; see Nasser et al. Nature (2021). Also on FTP (requires an FTP client): ftp://ftp.broadinstitute.org/outgoing/lincRNA/ABC/AllPredictions.AvgHiC.ABC0.015.minus150.ForABCPaperV3.txt.gz
ABC pipeline outputs for 131 cell types and tissues — see Nasser et al. Nature (2021). FTP (requires an FTP client): ftp://ftp.broadinstitute.org/outgoing/lincRNA/ABC/Nasser2021-Full-ABC-Output/
Web browser for ABC predictions in 131 biosamples (by Fritz Lekschas)
Curated CRISPR datasets for validating predictive models of enhancer-gene regulation — see Nasser et al. Nature (2021)

Protocols and Code:

Schematic of the Activity-by-Contact model: enhancer activity times enhancer–promoter contact predicts regulatory-element-to-gene connections across hundreds of cell types.

CRISPRi tiling and FlowFISH

CRISPRi tiling is a method to characterize the functions of noncoding elements on the expression of a gene of interest. It combines CRISPR interference (CRISPRi) with various gene-centric screening tools, including RNA FlowFISH.

We have applied these tools to dissect enhancer-promoter regulation across many genes and cell types.

Publications and Data:

Fulco and Nasser et al. Nature Genetics (2019). Description of the CRISPRi-FlowFISH method, and application to comprehensively test all putative regulatory elements around 30 genes in K562 cells.
Fulco et al. Science (2016). Initial description of the CRISPRi tiling approach, and application to tile gRNAs across ~1.2 Mb of sequence around two genes of interest (GATA1 and MYC).

Protocols:

CRISPRi tiling with FlowFISH: KRAB-dCas9 and a tiling gRNA library target DHS peaks, cells are stained by RNA FISH for the gene of interest, sorted by expression into six bins, and gRNAs sequenced to infer each element's effect.

Prime editing tools: Variant-EFFECTS and Swap-seq

Variant-EFFECTS uses pooled prime editing, flow sorting, and sequencing to install precise edits in regulatory DNA — single-nucleotide variants, motif insertions, or other designed changes — and measure each edit's quantitative effect on the expression of a gene of interest.

Swap-seq uses pooled twin prime editing to tile precise, barcoded deletions across a locus, mapping the enhancers, silencers, and CTCF sites that control a gene.

Publications and Data:

Martyn and Montgomery et al. Cell (2025). Description of the Variant-EFFECTS method, and application to measure the effects of variants and designed edits in enhancers and promoters. Data: GSE285157; models: Zenodo.
Cai et al. bioRxiv (2026). Description of the Swap-seq method, and application to map enhancers, silencers, and topological elements with prime editing deletion screens. Data: GSE348496, GSE347552.

Protocols and Code:

Variant-EFFECTS: designed edits are delivered to a target regulatory element and their effects on gene expression measured — used to mutagenize endogenous regulatory DNA, insert transcription factor motifs, benchmark sequence models, and tune gene expression up or down.

Swap-seq: tile prime-editing deletions around a gene of interest, induce editing in a pooled screen, sort cells by RNA FlowFISH, and sequence deletion-specific barcodes to quantify each deletion's effect on expression.

RNA Antisense Purification (RAP)

RNA Antisense Purification (RAP) is a biochemical purification method that allows unbiased identification of the DNA, RNA, or proteins associated with an RNA of interest. RAP involves crosslinking cells to fix RNA interactions, capturing the RNA of interest using long (60–120-nucleotide) antisense oligos, and identifying associated biomolecules using DNA sequencing, RNA sequencing, or proteomics.

We have applied these tools to discover how lncRNAs localize to genomic DNA using the 3D organization of the genome, and how RNA-RNA interactions can guide lncRNA localization to chromatin.

Publications and Data:

Engreitz et al. Cell (2014). Description of RAP-RNA, and updated RAP-DNA protocol. RAP-DNA and RAP-RNA for U1 snRNA, Malat1 lncRNA, and other small noncoding RNAs.
Engreitz et al. Science (2013). Initial description of the RAP-DNA protocol, and application to study how Xist spreads across the X chromosome.

RNA Antisense Purification workflow: crosslinked cells are lysed, biotinylated probes hybridize to the target RNA, and captured complexes are eluted for DNA sequencing, RNA sequencing, or mass spectrometry.

Protocols and Code:

RAP-DNA and RAP-RNA were developed together with the Mitch Guttman Lab.