Datasets:

Genentech
/

human-chromhmm-fullstack-data

license: mit
task_categories:
  - tabular-classification
tags:
  - biology
  - genomics
pretty_name: ChromHMM fullstack annotation of the human genome
size_categories:
  - 1M<n<10M

human-chromhmm-fullstack-data

Dataset Summary

This dataset provides a multi-class annotation of genomic regions across the hg38 genome. It is derived from the ChromHMM fullstack annotation (Vu & Ernst, 2022; https://doi.org/10.1186/s13059-021-02572-z). Genomic regions are classified into 16 chromatin states. The data is derived from https://public.hoffman2.idre.ucla.edu/ernst/2K9RS//full_stack/full_stack_annotation_public_release/hg38/hg38_genome_100_segments.bed.gz.

Repository Content

data.csv: The main dataset stored in comma-separated tabular format.
1_data.ipynb: Jupyter notebook containing the preprocessing steps used to generate the .csv file.

Dataset Structure

Rows: 5,809,104
Columns: 7

Data Splits

Split	Count
train	5,042,325
valid	364,387
test	402,392

Chromatin States (16 classes)

State	Count
Quies	1,485,576
Acet	639,669
EnhA	613,794
ReprPC	610,147
Tx	561,526
EnhWk	543,113
HET	521,161
TxWk	254,518
TxEnh	190,465
TxEx	121,833
PromF	88,429
GapArtf	51,474
BivProm	48,242
znf	34,146
TSS	24,402
DNase	20,609

Column Descriptions

Column	Type	Description
chrom	string	Chromosome name (e.g., chr1)
start	int	Start coordinate of the genomic interval
end	int	End coordinate of the genomic interval
state	string	Chromatin state annotation (16 classes)
interval_idx	int	Unique numerical index for the specific genomic interval
enformer_split	string	Overlap with the data splits used for training the Enformer model
split	string	Splits used for downstream modeling (train/valid/test)

Usage

import pandas as pd
from huggingface_hub import hf_hub_download

file_path = hf_hub_download(
    repo_id="Genentech/human-chromhmm-fullstack-data",
    filename="data.csv",
    repo_type="dataset"
)

df = pd.read_csv(file_path)