Atlas of Transcription Factor Binding Sites from ENCODE DNase Hypersensitivity Data Across 27 Tissue Types

Abstract

There is intense interest in mapping the tissue-specific binding sites of transcription factors in the human genome to reconstruct gene regulatory networks and predict functions for noncoding genetic variation. DNase-seq footprinting provides a means to predict the genome-wide binding sites for hundreds of transcription factors (TFs) simultaneously. However, despite the public availability of DNase-seq data for hundreds of samples, there is neither a unified analytical workflow nor a publicly accessible database providing the locations of footprints across all available samples. Here, we describe the implementation of a workflow for uniform processing of footprints using two state-of-the-art footprinting algorithms: Wellington and HINT. Our workflow then scans footprints for 1,530 sequence motifs to predict binding sites for 1,515 human transcription factors. We tested our workflow using 21 DNase-seq experiments of lymphoblastoid cell lines, generated by the ENCODE project. We trained a machine learning model to predict TF binding sites, integrating footprints with additional biologically-related features. This model achieved a maximum MCC of 0.423 and an AUC of 0.943 compared to ENCODE ChIP-seq data for 62 TFs in the same cell type. We applied our workflow to detect footprints in 206 DNase-seq experiments from ENCODE, spanning 27 human tissues. These footprints describe an expansive landscape of TF occupancy in the human genome. Across all tissues, we detected high-quality footprints spanning 9.8% of all nucleotides in the human genome with scores found to enrich for true positives. The highest tissue-specific coverage was observed for samples in the brain (4.4%), followed by extra-embryonic structure (2.6%) and skin (2.4%). In addition, we report a more lenient footprinting call set, providing some evidence of TF occupancy in at least one tissue for 34% of all genomic positions. Our cloud-based workflow and a database with all footprints and TF binding site predictions are available at www.trena.org.

Omics in Systems Biology: Current Progress and Future Outlook

“…Methods for studying DNA–protein interactions include electrophoretic mobility shift assays, chromatin immunoprecipitation (ChIP), DNase footprinting, and systematic evolution of ligands by exponential enrichment (SELEX). [ 58–61 ] Unfortunately, these methods are only useful if the DNA remains intact throughout its interaction with a protein. Protein binding microarrays have also been used to study transcription factor (TF)‐DNA interactions, however, the recognition sequences of many TFs are longer than those that can be coupled to the array.…”

Reproducible big data science: A case study in continuous FAIRness

“…BDBags a "bag of bags," a BDBag that contains references to a set of other BDBags. As the ENCODE data consist primarily of short sequence reads, Funk et al [13] ran 277 the sequence alignment process twice, with seed lengths of 16 and 20, respectively. 1 278 thus produces two BDBags per tissue type, for a total of 54.…”

“…The Funk et al [13] workflow uses encode2bag to create BDBags for each of the 27 254 tissue types in ENCODE, each with its own Minid. For example, the DNase-seq data 255 associated with adrenal tissue is at minid:b9w37t.…”

“…The input to 4 , shown as Motif in Fig 1, is a collection of 1,530 nonredundant TF 330 binding motifs assembled by Funk et al [13]. This motif collection was assembled by 331 using the Tomtom program from the MEME suite [36] to identify non-redundant motifs 332 within the JASPAR 2016 [37], HOCOMOCO v10 [38], UniPROBE [39], and 333 SwissRegulon [40] motif libraries, each of which was accessed via the Bioconductor R 334 package MotifDb [41].…”

“…Connect textual statements to underlying results. Our use of Minids would make it 410 easy for Funk et al [13] to reference specific data in their text. They do not do at 411 present, but may in a future version of their paper.…”