bedrockbio

Open-Access Computational Biology Datasets

Description

Efficiently access a curated library of open-access computational biology datasets. Tables support predicate pushdown and projection to the cloud storage backend, enabling quick, iterative access to otherwise massive, unwieldy tables.

bedrockbio consists of five user-facing functions:

• list_namespaces(): returns a character vector of available namespace (data source) identifiers
• describe_namespace("<name>"): returns metadata, citation, license, instructions, and the tables for a namespace
• list_tables(): returns a character vector of available table identifiers
• describe_table("<name>"): returns metadata, citation, partition and sort keys, and column definitions for a table
• load_table("<name>"): returns a lazily-evaluated data frame for a table

dplyr verbs (filter, select) can be used on the data frame returned by load_table to push down row filters and column selections to the storage backend. Filtering on the partition columns returned by describe_table gives the fastest reads.

Installation

Install from CRAN:

install.packages("bedrockbio")

Or install the current development version from GitHub:

# install.packages("pak")
pak::pak("bedrock-bio/bedrock-bio/r")

Examples

Load the package (and dplyr for downstream data frame manipulation):

library(bedrockbio)
library(dplyr)

List available tables:

list_tables()

Describe a table to see its metadata, citation, and columns:

describe_table("ukb_ppp.pqtls")

Lazily load a table, filter on partition columns (for fastest reads), select columns, and collect the relevant subset into an in-memory data frame:

df <- load_table("ukb_ppp.pqtls") |>
  filter(
    ancestry == "EUR",
    protein_id == "A0FGR8",
    panel == "Inflammation"
  ) |>
  select(
    chromosome,
    position,
    effect_allele,
    other_allele,
    beta,
    neg_log_10_p_value
  ) |>
  collect()

Dataset Requests

To request the addition of a new table to the library, open an issue.