Predicting transcription factor binding in single cells through deep learning

Abstract

Characterizing genome-wide binding profiles of transcription factors (TFs) is essential for understanding biological processes. Although techniques have been developed to assess binding profiles within a population of cells, determining them at a single-cell level remains elusive. Here, we report scFAN (single-cell factor analysis network), a deep learning model that predicts genome-wide TF binding profiles in individual cells. scFAN is pretrained on genome-wide bulk assay for transposase-accessible chromatin sequencing (ATAC-seq), DNA sequence, and chromatin immunoprecipitation sequencing (ChIP-seq) data and uses single-cell ATAC-seq to predict TF binding in individual cells. We demonstrate the efficacy of scFAN by both studying sequence motifs enriched within predicted binding peaks and using predicted TFs for discovering cell types. We develop a new metric “TF activity score” to characterize each cell and show that activity scores can reliably capture cell identities. scFAN allows us to discover and study cellular identities and heterogeneity based on chromatin accessibility profiles.

Single-cell multi-omics of human clonal hematopoiesis reveals that DNMT3A R882 mutations perturb early progenitor states through selective hypomethylation

“…While Dnmt3a R878H HSPCs displayed a more modest increase in chromatin accessibility, this may be due to the global open chromatin in stem cells reducing the ability to measure specific enrichments 137,138 . Overall, as chromatin accessibility has been demonstrated to accurately reflect TF activity 136 , these data provided further evidence for the model in which the DNMT3A mutation enhances the activity of TFs whose binding motifs are prone to hypomethylation through enrichment in the hypomethylated sequence motif. This model then provides the basis of enhanced MYC/MAX target gene expression in the DNMT3A mutated cells observed in the GoT data (Fig.…”

“…While recent progress has been made in single-cell chromatin binding assays [133][134][135] , the ability to determine the weaker signal of TF binding in single cells remains a challenge. We therefore performed a chromatin accessibility assay, shown to be a reliable surrogate for determining TF activity 136 , on single nuclei (n = 46,496 cells, Fig. 5d, Extended Data Fig.…”

Deep Learning Applications in Single-Cell Omics Data Analysis

“…In a recent study, Fu et al (Fu, Zhang et al 2020) introduced scFAN (Single Cell Factor Analysis Network), a DL model for determining genome-wide TF binding profiles in individual cells. The scFAN pipeline consists of a "pre-trained model" trained on bulk data and then used to predict TF binding at the cellular level using DNA sequence, aggregated associated scATAC-seq data, and mappability data.…”

“…They suggested a novel metric called "TF activity score" to classify each cell and demonstrated that the activity scores could accurately capture cell identity. Generally, scFAN is capable of connecting open chromatin states with transcript factor binding activity in individual cells, which is beneficial for a deeper understanding of regulatory and cellular dynamics (Fu, Zhang et al 2020).…”

Accelerating Identification of Chromatin Accessibility from noisy ATAC-seq Data using Modern CPUs

“…Given the exponentially increasing throughput and reducing cost of next-generation sequencing and the corresponding rise in ATAC-seq experiments, it is critical to develop accurate and high throughput methods for denoising ATAC-seq data and peak calling. To this end, several deep learning based methods have been proposed to perform denoising and peak detection from 1D sequencing data [1][2][3][4][5][6][7][8]. These deep learning based methods typically use deep convolutional neural networks (CNNs).…”