Robust footprinting with sample-specific Tn5 bias correction for bulk and single cell ATAC-seq.

Abstract

Accurate detection of transcription factor (TF) and nucleosome occupancy from assay for transposase-accessible chromatin via sequencing (ATAC-seq) remains challenging due to sequence-dependent Tn5 cleavage bias. We show that this cleavage bias varies across samples and introduce TraceBind, an ATAC-seq footprinting framework that performs sample-specific Tn5 bias correction by fine-tuning a pretrained cleavage model using mitochondrial DNA reads, followed by multiscale detection with empirical false discovery rate control. Across bulk and single-cell ATAC-seq datasets, with validation by naked DNA controls, matched ChIP-seq and CUT&RUN experiments, and TF degradation perturbations, TraceBind reduces false positives while maintaining high sensitivity compared to existing methods. TraceBind enables improved downstream single-cell analyses, including transcription factor activity inference and cell age prediction. Base-resolution comparisons with DNA foundation model-based sequence annotations show enriched overlap but reveals many high-confidence footprints not captured by DNA foundation models, highlighting complementary regulatory information beyond what can be learned from DNA sequence alone.