TF Footprint & Differential Footprint Analysis (Orchestrator)

Purpose

Identify TF binding sites from ATAC-seq footprints and compare TF occupancy across conditions. Footprinting detects physical DNA protection by bound TFs within accessible chromatin, complementing motif accessibility (ChromVAR) with occupancy evidence.

Key distinction:

ChromVAR measures accessibility (potential for binding — chromatin open at motif sites)
Footprinting measures occupancy (actual binding — protein physically protects DNA from Tn5)

Upstream: Peak calling (MACS2/MACS3), pseudobulk BAM generation (ArchR/Signac), ChromVAR deviations Downstream: TF regulatory networks, quadrant analysis with ChromVAR, GRN inference (SCENIC+)

Tool Selection Guide

Feature	TOBIAS (`tobias-footprint-bindetect`)	HINT-ATAC (`hint-atac-differential-footprint`)	Signac (`signac-footprint-visualization`)
Algorithm	Tn5 bias correction (linear regression) + footprint scoring	HMM-based detection (learns footprint states)	Normalized Tn5 insertion profiles
Tn5 Bias Correction	Best in class (`ATACorrect`)	Good (bias table)	Basic (k-mer model)
Differential Analysis	Excellent (pairwise; p-values via `BINDetect`)	Good (multi-condition; `rgt-hint differential`)	Weak (visualization only)
Pseudobulk Support	High (designed for merged BAMs)	High (designed for merged BAMs)	Medium (fragment files, slow at scale)
Output	BigWigs, volcano plots, TSV tables	BED files, line plots, statistics TXT	Seurat objects, ggplot2 figures
Best For	Quantitative pairwise differential	Multi-condition overview heatmaps	Publication figures

Default routing:

Quantitative pairwise differential (WT vs KO with p-values) → tobias-footprint-bindetect
Multi-condition overview (3+ conditions in one run) → hint-atac-differential-footprint
Publication figures for selected TFs → signac-footprint-visualization
Per-cell TF activity (not footprinting) → chromvar-motif-accessibility
Combined approach (recommended): TOBIAS for quantitation + Signac for figures + HINT-ATAC if >2 conditions

Shared Motif Database Recommendations

For Mouse (mm10)

Database	Version	Mouse TFs	DC Coverage	Format	Notes
JASPAR	2024 Core Vertebrates	~800+	Excellent (IRF8, BATF::JUN, SPI1, ETS1, RUNX)	`.jaspar`, `.meme`	Field standard, recommended
HOCOMOCO	v11	~400 mouse	Good (IRF, ETS, RUNX families)	`.meme`, `.pcm`	High quality, curated
CIS-BP	2.0	~1200+	Comprehensive	Custom TSV	Broadest, some redundancy
chromVARmotifs	v1	Curated	Good	R `PWMatrixList`	Used by chromVAR natively

Recommendation: Use JASPAR 2024 Core Vertebrates as the shared motif set across TOBIAS, HINT-ATAC, Signac, and ChromVAR. Different PWMs for the same TF yield different binding-site calls and break cross-tool correlation.

Exporting ChromVAR motifs to MEME (for TOBIAS reuse)

library(TFBSTools)
library(universalmotif)

motif_pwms <- GetMotifData(obj, slot = "pwm")  # From Signac object
motif_list <- lapply(names(motif_pwms), function(nm) {
  convert_motifs(motif_pwms[[nm]], class = "universalmotif-universalmotif")
})
write_meme(motif_list, file = "chromvar_motifs.meme", overwrite = TRUE)

Quadrant Analysis (Footprint × Accessibility)

ChromVAR (accessibility) and TOBIAS (occupancy) provide complementary evidence. Integrate via scatter plot:

Quadrant	ChromVAR	Footprint	Interpretation
Q1	High delta	High delta	True drivers — chromatin open AND TF bound
Q2	High delta	Low delta	Poised / pioneer — accessible but TF not occupying (short residence, indirect effect)
Q3	Low delta	High delta	Rare — binding in less accessible regions (pioneer) or artifact
Q4	Low delta	Low delta	Not active at these sites

Implementation in R

library(tidyverse)

# Load TOBIAS results
tobias <- read_tsv("differential/cDC1_WT_vs_KO/bindetect_results.txt")

# Load ChromVAR differential results (delta deviation per motif)
chromvar_diff <- read_csv("hub/A3_tf_activity_by_celltype.csv")

# Merge on TF/motif name (requires name harmonization across DBs)
combined <- inner_join(
  tobias %>% select(motif_id = output_prefix,
                    tobias_score = WT_Batf3_KO_change,
                    tobias_pval  = WT_Batf3_KO_pvalue),
  chromvar_diff %>% select(motif_id, chromvar_delta = delta_deviation),
  by = "motif_id"
)

ggplot(combined, aes(x = chromvar_delta, y = tobias_score)) +
  geom_point(aes(color = tobias_pval < 0.05), alpha = 0.6) +
  geom_hline(yintercept = 0, linetype = "dashed") +
  geom_vline(xintercept = 0, linetype = "dashed") +
  ggrepel::geom_text_repel(
    data = combined %>% filter(tobias_pval < 0.05),
    aes(label = motif_id), size = 3, max.overlaps = 20
  ) +
  scale_color_manual(values = c("grey70", "red3")) +
  labs(x = "ChromVAR Delta Deviation (Accessibility)",
       y = "TOBIAS Differential Binding Score (Occupancy)",
       title = "TF Activity: Accessibility vs Occupancy") +
  theme_minimal()

Cross-Tool Pitfalls

Issue	Cause	Fix
ChromVAR vs TOBIAS disagree	Different motif databases	Use identical PWM set for both (see export snippet above)
Motif name mismatch when joining	Database-specific IDs (`MA0080.5` vs `SPI1`)	Build an explicit ID-to-symbol mapping table
TOBIAS / HINT-ATAC contradict	Pairwise vs multi-condition framing	Compare pairwise sub-contrasts within HINT-ATAC's multi-condition run

Resources

TOBIAS: Bentsen et al., Nat Commun 2020 (DOI: 10.1038/s41467-020-18035-1) — https://github.com/loosolab/TOBIAS
HINT-ATAC: Li et al., Genome Biol 2019 (DOI: 10.1186/s13059-019-1642-2) — https://reg-gen.readthedocs.io/
Signac footprint vignette: https://stuartlab.org/signac/articles/footprint
JASPAR 2024: https://jaspar.elixir.no/
HOCOMOCO v11: https://hocomoco11.autosome.org/

When not to use

Do not use for single-cell resolution TF activity — use chromvar-motif-accessibility.
Do not use for motif enrichment on DARs — use pycistarget-motif-enrichment.

tf-footprint-differential-analysis

SKILL.mdskills/tf-footprint-differential-analysis/SKILL.md