CT-FOCS: a novel method for inferring cell type-specific enhancer-promoter maps

Abstract

Spatiotemporal gene expression patterns are governed to a large extent by enhancer elements, typically located distally from their target genes. Identification of enhancerpromoter (EP) links that are specific and functional in individual cell types is a key challenge in understanding gene regulation. We introduce CT-FOCS, a new statistical inference method that utilizes multiple replicates per cell type to infer cell type-specific EP links. Computationally predicted EP links are usually benchmarked against experimentally determined chromatin interaction measured by ChIA-PET. We expand this validation scheme by introducing the concept of connected loop set, which combines loops that overlap in their anchor sites. Analzying 1,366 samples from ENCODE, Roadmap epigenomics and FANTOM5, CT-FOCS inferred highly cell type-specific EP links more accurately than a state-of-the-art method. We illustrate how our inferred EP links drive cell type-specific gene expression and regulation.TargetFinder [13]. All these methods rely on data of multiple chromatin marks and expression data for the studied cell types.The JEME algorithm finds global and cell type-active EP links (but not necessarily cell typespecific) using only 1-5 different omics data types [14]. Each reported EP link is given a score denoting tendency to be active in a given cell type. JEME reports an average of 4,095 active EP links per cell type, and most of these may be nonspecific.Several recent studies aimed at finding ct-links experimentally. Rajarajan et al. [15] used insitu HiC and schizophrenia risk locus to identify 1,702 and 442 neuronal progenitor cell (NPC) specific and neuron specific 3D chromatin interactions for 386 and 385 genes, respectively. Some of the NPC and neuron-specific interactions may be enhancer-promoter interactions (or ct-links). Gasperini et al. [16] used CRISPR screening to perturb 5,920 human candidate enhancers that may affect gene expression at the single-cell level in combination with eQTL analysis, and identified 664 EP links covering 479 genes enriched with K562-specific genes and lineage-specific transcription factors (TFs; reviewed in [17]). Remarkably, both studies reported far fewer links than JEME, indicating that only a small portion of EP links that are active in a cell type are specific for it.Here, we introduce CT-FOCS (Cell Type FDR-corrected OLS with Cross-validation and Shrinkage), a novel method for inferring ct-links from large-scale compendia of hundreds of cell types measured by a single omic technique (e.g., DNase Hypersensitive Sites sequencing; DHS-seq). It is built upon our previously published method, FOCS [18], which infers global EP links that show high correlation between the enhancer and the promoter activity patterns across many samples. Given the omic profile for a set of cell types, each one with replicates, CT-FOCS uses linear mixed effect models (LMMs) to infer ct-links. CT-FOCS was applied on public DNase Hypersensitive Sites (DHS) profiles from ENCODE and Roadmap Epigenomics [19][20][21], and ca...