BrentLab/barkai_compendium

Barkai Compendium

This collects the ChEC-seq data from the following GEO series:

• —GSE179430
• —GSE209631
• —GSE222268

The metadata for each is parsed out from the SraRunTable, or in the case of GSE222268, the NCBI series matrix file (the genotype isn't in the SraRunTable)

The Barkai lab refers to this set as their binding compendium.

The genotypes for GSE222268 are not clear enough to me currently to parse well.

Accessing Data

The examples below require the HuggingFace Hub client (pip install huggingface_hub).

Direct parquet access

The repository is large and contains both single parquet files (metadata) and partitioned datasets (genome_map coverage data). Download the metadata files first to identify which partitions are relevant before fetching coverage data.

Single parquet file example (metadata):

from huggingface_hub import snapshot_download
import duckdb

repo_path = snapshot_download(
    repo_id="BrentLab/barkai_compendium",
    repo_type="dataset",
    allow_patterns="*metadata.parquet",
)
conn = duckdb.connect()
# returns a pandas DataFrame with the first 5 rows
conn.execute(
    "SELECT * FROM read_parquet(?) LIMIT 5",
    [f"{repo_path}/GSE178430_metadata.parquet"],
).df()

Partitioned dataset example (genome_map coverage):

repo_path = snapshot_download(
    repo_id="BrentLab/barkai_compendium",
    repo_type="dataset",
    allow_patterns="genome_map/series=GSE179430/accession=GSM5417602/*.parquet",
)
conn.execute(
    "SELECT * FROM read_parquet(?) LIMIT 5",
    [f"{repo_path}/genome_map/**/*.parquet"],
).df()

Accessing using R

Clone the repository and read parquet files directly with arrow:

# install.packages("arrow")
arrow::read_parquet("GSE178430_metadata.parquet")