DeFCoM

DeFCoM identifies transcription factor binding sites by analyzing genome-wide chromatin accessibility data (DNase-seq and ATAC-seq) using a supervised learning framework that models site-to-site variability in footprint signals.

Key Features:

• Input data: Uses genome-wide DNase-seq and ATAC-seq chromatin accessibility data to detect transcription factor footprints.
• Supervised learning: Employs a supervised learning framework to classify bound versus unbound motif sites for specific transcription factors.
• Variability modeling: Explicitly models site-to-site variability in footprint signal profiles rather than assuming homogeneity.
• Training data: Trained on DNase-seq data from the ENCODE project.
• Benchmarking: Evaluated against nine existing footprinting methods across four human cell lines and eighteen transcription factors.
• Performance: Demonstrates improved accuracy over compared methods in identifying bound motif sites across the evaluation sets.
• Cross-assay robustness: Detects footprints with similar accuracy on ATAC-seq data as on DNase-seq data.
• Factor analysis: Provides analyses of the impact of biological and technical factors on footprint prediction quality.
• Implementation: Implemented in Python.

Scientific Applications:

• Transcription factor binding mapping: Pinpoints genomic locations of transcription factor interactions using footprint patterns from DNase-seq and ATAC-seq.
• Gene regulation studies: Supports investigation of transcriptional regulation dynamics across cell types and conditions.
• Method benchmarking: Enables comparative evaluation of footprinting algorithms using standardized benchmarks across cell lines and transcription factors.
• Assay comparison: Facilitates assessment of footprint detectability and consistency between DNase-seq and ATAC-seq datasets.

Methodology:

Uses a supervised learning model trained on ENCODE DNase-seq data that explicitly models site-to-site variability in footprint signals, evaluated against nine existing footprinting methods across four human cell lines and eighteen transcription factors, and applied to ATAC-seq data; implemented in Python.