Multi-scale Footprinting: Tn5 Bias and Transcription Factor Binding Models

Pre-trained convolutional neural network models for predicting Tn5 transposase insertion bias from local DNA sequence context. The data includes genome-wide bias predictions for multiple reference genomes (hg38, hg19, mm10, etc.), dispersion models for footprinting, and transcription factor binding prediction models. Yan Hu from Harvard University Press released this data associated with the multi-scale footprinting project.

Use Cases

• Correcting Tn5 insertion bias in ATAC-seq data based on the provided pre-computed genome-wide bias files.
• Predicting transcription factor binding sites using the provided footprint-to-TF binding prediction models.
• Clustering transcription factors based on multi-scale footprint patterns using the provided cluster label files.
• Following tutorials for analyzing bulk ATAC-seq data from PBMC samples using the provided example fragment files and result bigwigs.

Strengths

• Includes pre-trained models for Tn5 bias prediction and TF binding prediction, which can save computational time.
• Provides pre-computed bias tracks for nine different reference genomes, including human (hg38, hg19), mouse (mm10, mm39), and others.
• Contains specific tutorial data for PBMC bulk ATAC-seq analysis, including filtered peaks and example result files.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description notes that one TF binding prediction model (TFBS_model.h5) is obsolete, directing users to a different method.

Provenance

Source

paperswithcode

Collection Method

Data associated with a research project; includes model outputs, pre-computed files, and tutorial materials.