TF Footprinting - Usage Guide

Overview

Detect transcription factor binding footprints in ATAC-seq data: short DNA stretches of reduced Tn5 cleavage within accessible regions where bound TFs physically protect DNA. Covers Tn5 sequence-bias correction (the single most important step), per-base footprint scoring, and motif-anchored bound/unbound classification. Compares TOBIAS, HINT-ATAC, Wellington, PIQ, and scprinter; documents per-TF-family failure modes (CTCF clean, nuclear receptors transient, ZBTBs unfootprintable).

Prerequisites

conda install -c bioconda tobias rgt pydnase samtools bedtools   # HINT-ATAC ships in the `rgt` package (invoked as `rgt-hint`)
pip install pyBigWig    # scPrinter installs from source: git clone https://github.com/buenrostrolab/scPrinter && cd scPrinter && pip install ./

Inputs required: - Deduplicated, MAPQ-filtered, chrM-stripped BAM (>= 50M nuclear reads recommended) - Consensus peakset (BED or narrowPeak) - Reference genome FASTA (matched build to BAM) - ENCODE blacklist - Motif database: JASPAR 2024 CORE vertebrates PFM (default) or HOCOMOCO v12

wget https://jaspar.genereg.net/download/data/2024/CORE/JASPAR2024_CORE_vertebrates_non-redundant_pfms.txt
mv JASPAR2024_CORE_vertebrates_non-redundant_pfms.txt JASPAR2024.pfm

Quick Start

Tell your AI agent what you want to do: - "Run the TOBIAS three-step pipeline (ATACorrect, ScoreBigwig, BINDetect) on two conditions" - "Verify Tn5 bias correction worked by plotting CTCF aggregate footprint" - "Use scprinter for multi-scale footprinting that handles both CTCF and nuclear receptors" - "Filter to NFR fragments (< 100 bp) before footprinting" - "Cross-validate predicted FOXA1 footprints against published ChIP-seq peaks"

Example Prompts

Differential TOBIAS

"Run TOBIAS ATACorrect on each condition's BAM, compute footprint scores per condition, then BINDetect with --cond_names treated control to get per-TF differential bound counts and p-values."

Single-Tool Sanity Check

"Plot aggregate footprint at JASPAR CTCF MA0139.1 motif sites using TOBIAS PlotAggregate. Confirm clean V-shape -- if shallow, depth or bias correction failed."

Multi-Scale TF Activity

"Run scprinter on this 80M-read library to get multi-scale footprints; compare CTCF (long footprint) and FOXA1 (short pioneer footprint) activity."

Single-Cell Footprinting

"Run scprinter on the scATAC fragments file with cell-type clusters; report per-cluster TF activity for hematopoietic lineage TFs (PU.1, GATA1, KLF1)."

Tool Reconciliation

"Run both TOBIAS and HINT-ATAC; report only TFs called bound by both tools at >= 50% site overlap as high-confidence."

NFR-Only Pipeline

"Filter BAM to fragments < 100 bp, then run TOBIAS three-step on the NFR BAM; compare to full-fragment BAM to see footprint sharpness gain."

What the Agent Will Do

Verify input depth (>= 50M nuclear reads) and library QC
Optionally filter BAM to NFR fragments for sharper footprints
Run Tn5 bias correction (TOBIAS ATACorrect or HINT's built-in dinucleotide model)
Validate bias correction by checking aggregate at CTCF (should show clean V)
Score per-base footprints across consensus peakset
Classify motif sites as bound/unbound with motif database (JASPAR/HOCOMOCO)
For differential: run BINDetect with --cond_names; report per-TF differential scores and p-values
Cross-validate against ChIP-seq peaks where available (especially for nuclear receptors and pioneer TFs)
Report per-TF-family caveats (transient binding, asymmetric pioneer footprints, dynamic ZBTBs)

Tool Decision Quick Reference

Goal	Tool
Standard differential TF activity, two conditions	TOBIAS three-step
Single condition, find bound TFs	TOBIAS or HINT-ATAC
Multi-scale (CTCF + nuclear receptors simultaneously)	scprinter
Single-cell ATAC	scprinter
Plant / non-model	TOBIAS with custom motif database (CIS-BP)
Lower depth (< 50M)	PIQ (less depth-sensitive); or pool replicates first

TF Family Quick Reference

Family	Footprint quality	Notes
CTCF	Excellent (gold standard)	Use for QC validation of bias correction
AP-1 (FOS/JUN)	Good but composite	Use HOCOMOCO for specific dimers
Pioneer (FOXA1, GATA, OCT4)	Asymmetric	Stranded mode in HINT-ATAC; expect one-shoulder V
Nuclear receptors (ER/AR/GR)	Shallow / transient	Validate with ChIP-seq; pool replicates
Forkhead/Homeobox short motifs	At resolution limit	scprinter multi-scale; aggregate over many sites
ZBTB / dynamic TFs	Often unfootprintable	Use ChIP-seq instead

Tn5 Bias Correction Quick Reference

Method	Tool	Pros	Cons
+/-12 bp k-mer window, DWM	TOBIAS ATACorrect	Default; well-validated	Genome-wide; single-process
Dinucleotide HMM	HINT-ATAC	Built into HINT pipeline	Less control over output
seqOutBias	seqOutBias	Independent of footprinting tool	Manual integration
Empirical from naked DNA	(research-grade)	Most accurate	Requires input library

Tips

Tn5 bias correction is mandatory. Skipping it produces "footprints" reflecting Tn5 sequence preference, not TF binding.
Validate bias correction by inspecting the aggregate footprint at CTCF motifs first. Clean V-shape means correction worked; inverted-V or flat means it failed.
50M nuclear reads is the practical floor; below that, weak / transient binding TFs cannot be reliably called.
For nuclear receptors, footprint absence does not mean absence of binding. Validate with ChIP-seq.
ATAC footprints DO NOT replicate ChIP-seq for all TFs. Concordance is good for stable binders (CTCF ~70%), poor for transient (nuclear receptors ~30%).
Pioneer factor footprints are asymmetric (one face of DNA on histone). HINT-ATAC has stranded mode for this.
For differential, identical peakset and identical blacklist must be used per condition; otherwise differential scores are not comparable.
BINDetect differential score magnitudes 0.1-0.5 are typical for biologically relevant changes; > 1.0 is rare and warrants investigation.
JASPAR is conservatively curated; HOCOMOCO has more motifs with quality grades; CIS-BP for non-model organisms.
NFR-only filtering (< 100 bp fragments) sharpens footprints but discards mononucleosome-borne information; do it for footprinting only, not for general accessibility.
TOBIAS PlotAggregate is the most useful diagnostic; always run it for top-candidate TFs.

atac-seq/atac-peak-calling - Generate consensus peakset; NFR peaks are sharper for footprinting
atac-seq/atac-qc - Confirm depth >= 50M and TSS enrichment before footprinting
atac-seq/motif-deviation - Complementary chromVAR for accessibility variability (different question)
atac-seq/single-cell-atac - scprinter for single-cell footprinting
chip-seq/peak-annotation - ChIP-seq cross-validation
sequence-manipulation/motif-search - Underlying motif-scanning details
gene-regulatory-networks/scenic-regulons - Downstream TF -> gene regulatory networks