adamyhe/procap-atlas-tracks

PRO-cap Atlas Tracks

This dataset hosts browser-ready track assets for the ENCODE PRO-cap atlas on the human GRCh38/hg38 genome, as well as the processed training/evaluation data for the PRO-cap model atlas. It contains observed strand-specific PRO-cap signal BigWigs, processed TSS peak files, bigBed peak tracks, BPNet profile/count contribution score BigWigs, and UCSC track hub configuration files.

The track hub can be loaded in the UCSC Genome Browser:

https://huggingface.co/datasets/adamyhe/procap-atlas-tracks/resolve/main/ucsc/hub.txt

This repository is intended to be used together with:

• BPNet models: https://huggingface.co/adamyhe/procap-atlas
• Atlas metadata: https://huggingface.co/datasets/adamyhe/procap-atlas-metadata
• Code repository: https://github.com/kundajelab/procap-atlas

Dataset Details

• Curated by: Adam Y. He and collaborators
• Source project: ENCODE PRO-cap atlas
• Assay: PRO-cap
• Organism: Homo sapiens
• Genome assembly: GRCh38/hg38
• Number of experiments: 224
• Number of biosamples: 126
• License: MIT
• Dataset repo: https://huggingface.co/datasets/adamyhe/procap-atlas-tracks
• Code repository: https://github.com/kundajelab/procap-atlas
• Metadata repo: https://huggingface.co/datasets/adamyhe/procap-atlas-metadata

Uses

Direct Use

Use this dataset to:

• visualize ENCODE PRO-cap signal, peaks, and BPNet contribution scores in UCSC Genome Browser
• download processed strand-specific PRO-cap BigWigs for atlas experiments
• download processed peak BED.gz files and converted bigBed peak tracks
• inspect BPNet profile and count contribution score tracks alongside observed signal
• support downstream regulatory genomics analyses that need processed browser-scale track files

Out-of-Scope Use

This dataset does not contain raw sequencing reads, raw ENCODE analysis files, trained model checkpoints, or full experiment metadata tables. Model checkpoints are hosted in adamyhe/procap-atlas, and metadata/configuration files are hosted in adamyhe/procap-atlas-metadata. These tracks are research resources and should not be used for clinical or diagnostic decision-making.

Dataset Structure

observed/
+-- {exp_id}_pl.bigWig
+-- {exp_id}_mn.bigWig

peaks/
+-- bed/{exp_id}_{biosample}.bed.gz
+-- bigbed/{exp_id}_{biosample}_peaks.bb

attributions/
+-- bpnet/{exp_id}_profile.bigWig
+-- bpnet/{exp_id}_count.bigWig

ucsc/
+-- hub.txt
+-- genomes.txt
+-- hg38/trackDb.txt

Observed Signal

observed/{exp_id}_pl.bigWig and observed/{exp_id}_mn.bigWig contain processed PRO-cap signal for each ENCODE experiment. Plus-strand values are positive. Minus-strand values are stored as negative values so they appear below the axis in genome browsers.

Peaks

peaks/bed/ contains processed peak BED.gz files. peaks/bigbed/ contains UCSC-compatible bigBed conversions of those peaks. Peak files merge bidirectional and unidirectional TSS peak calls when both are available.

The merged 8-column peak format is:

Column	Field	Description
1	chrom	Chromosome
2	chromStart	Start position
3	chromEnd	End position
4	strand	+, -, or Both
5	confidence	Bidirectional confidence label; . for unidirectional peaks
6	peakType	Unidirectional or Bidirectional
7	summitsPos	Plus-strand summit position(s), or .
8	summitsMn	Minus-strand summit position(s), or .

BPNet Contribution Scores

attributions/bpnet/{exp_id}_profile.bigWig and attributions/bpnet/{exp_id}_count.bigWig contain BPNet profile and count contribution score tracks. In the UCSC hub these are displayed as standalone dynseq logo tracks with logo on.

UCSC Track Hub

The ucsc/ directory contains the track hub configuration. Experiments are grouped into supertracks by biosample. For capped experiments, observed signal, peaks, BPNet profile contribution scores, and BPNet count contribution scores are visible by default. Uncapped-library experiments are hidden by default.

Dataset Creation

Source Data

The source experiments are ENCODE PRO-cap datasets. Observed signal tracks and peak calls are downloaded from ENCODE using curated manifests in the companion procap-atlas repository.

Processing

The companion repository preprocesses tracks as follows:

• replicate BigWigs are merged per experiment while keeping plus and minus strands separate
• bidirectional and unidirectional peak files are merged into processed peak BED.gz files
• peak BED.gz files are converted to bigBed for UCSC browser compatibility
• BPNet contribution score NPZ files are converted to observed-nucleotide BigWigs and averaged over overlapping windows
• src/hub/generate_hub.py writes ucsc/hub.txt, ucsc/genomes.txt, and ucsc/hg38/trackDb.txt

Track assets are uploaded with src/hub/upload_tracks_hf.py, which stages a temporary repo-shaped directory and uses Hugging Face upload_large_folder.

Bias, Risks, and Limitations

• ENCODE biosample coverage is uneven across cell types, tissues, disease states, perturbations, and library construction types.
• Some experiments are perturbation, treatment, or uncapped-library datasets. Consult the companion metadata repo before pooling experiments or interpreting tracks as baseline cell-state data.
• The tracks are aligned to GRCh38/hg38 and should not be used directly on other genome assemblies.
• BPNet contribution score tracks reflect trained model behavior and should be interpreted alongside model performance metrics and experiment quality flags.
• Large BigWig and bigBed files require tools that support HTTP byte-range requests for efficient remote access.

How to Use

Load the hub in UCSC Genome Browser through My Data -> Track Hubs -> My Hubs:

https://huggingface.co/datasets/adamyhe/procap-atlas-tracks/resolve/main/ucsc/hub.txt

Download individual files directly from Hugging Face resolve URLs, for example:

https://huggingface.co/datasets/adamyhe/procap-atlas-tracks/resolve/main/observed/ENCSR083AMN_pl.bigWig
https://huggingface.co/datasets/adamyhe/procap-atlas-tracks/resolve/main/peaks/bigbed/ENCSR083AMN_B_cell_peaks.bb
https://huggingface.co/datasets/adamyhe/procap-atlas-tracks/resolve/main/attributions/bpnet/ENCSR083AMN_profile.bigWig

Programmatic users should pair these tracks with configs/experiment_config.yaml and configs/model_warning_flags.tsv from adamyhe/procap-atlas-metadata.

Citation

If you use these tracks, please cite the PRO-cap atlas repository and the underlying ENCODE experiments (Shah et al., 2025). For model-derived contribution score tracks, also cite the companion BPNet model repository and relevant attribution tooling.

Contact

For questions, bug reports, or reuse notes, please use the GitHub repository issues: https://github.com/kundajelab/procap-atlas/issues