biodata-cache
biodata-cache is a set of one-line functions that handle the entire process of caching and retrieving data (and metadata) from AIND data assets.
In the background, the cache repackages data/metadata into dataframes and stores them on S3 in versioned folders (data-asset-cache/bdc-v{version}/), or in memory for testing. Each release writes to its own versioned folder, so older versions of the website remain accessible while new versions are deployed. A top-level data-asset-cache/cache_versions.json index lists all available version folders.
Important: this package is not at 1.0. It is changing fast and breaking changes are still occurring, although rarely. To reduce the chance of impact on your code the cache tables are versioned. This does mean that if you want the latest version of the tables you need to keep biodata-cache up-to-date, but it also means your code won't immediately break when I change the way the tables work.
Installation
Note that you must set the backend to S3 or biodata-cache will automatically re-cache the tables locally in memory. This can take a LONG time.
pip install biodata-cache
export BIODATA_CACHE_BACKEND='S3'
Usage
Set backend
export BIODATA_CACHE_BACKEND='S3'
Options are 'S3', 'MEMORY'.
Fetch data
from biodata_cache import unique_project_names
project_names = unique_project_names()
Cache tables
Use get_cache_registry() to see all available cache tables and their metadata (descriptions, S3 paths, columns, etc.) for the installed version:
from biodata_cache import get_cache_registry
registry = get_cache_registry()
Use get_cache_versions() to list all available version folders across all deployed releases:
from biodata_cache import get_cache_versions
versions = get_cache_versions()
The per-version cache_registry.json lives at s3://allen-data-views/data-asset-cache/bdc-v{version}/cache_registry.json. The top-level index s3://allen-data-views/data-asset-cache/cache_versions.json lists all available version folders as a JSON array.
Hive-partitioned tables use key=value directory segments, enabling DuckDB queries like:
import duckdb
duckdb.query("""
SELECT * FROM read_parquet(
's3://allen-data-views/data-asset-cache/bdc-v0.27.3/qc/data.pqt',
hive_partitioning=true,
union_by_name=true
)
""")
The raw_to_derived function is not a table stored in S3, instead it is used by passing an asset_name (or list of asset names) and a modality. The function returns the latest derived asset matching the requested pattern.
Cells across every asset
Three tables answer "what do we know about every single cell in every data
asset?", joined on a stable cell_key:
| Table | Grain | Partitioned by | Holds |
|---|---|---|---|
cell_index |
one row per cell | — | Identity and provenance: asset, subject, acquisition, modality, probe/plane, source unit/ROI id |
cell_properties |
one row per cell | asset_name |
CCF location, mean rate, QC, cell type, and other measurements — wide and deliberately sparse |
cell_genes |
one row per genotyped cell | subject_id |
Transcriptomic cluster labels and gene counts |
import duckdb
from biodata_cache import cell_index, cell_properties
cells = cell_index() # every cell, one small fetch
props = cell_properties(asset_name=cells["asset_name"].iloc[0])
Most cells have only a handful of properties and NULL for the rest; that is the
expected shape. A sync run skips cell_properties partitions that already exist,
so pass force_rewrite=True if you need existing partitions rebuilt. cell_key is a hash of (asset_name, container, cell_ref), so it
is stable across cache rebuilds and safe to persist. cell_ref is the source
NWB's own unit/ROI identifier, so any row can be traced back to its source.
Custom cache table
The custom function allows you to store and retrieve your own user-defined DataFrames in the cache by name. This requires write authentication to the active backend.
from biodata_cache import custom
import pandas as pd
df = pd.DataFrame({"col": [1, 2, 3]})
custom("my_data", df)
retrieved_df = custom("my_data")
Update all cache tables
The cache is rebuilt on Code Ocean by a set of per-table sync jobs wired into a
Nextflow pipeline (asset_basics first, then the rest in parallel, then cell-by-everything).
Each job is
selected by the BIODATA_CACHE_SYNC_JOB environment variable and writes its own
registry fragment as it completes. See PIPELINE.md for the job
list, pipeline layout, and the version-bump/re-run procedure.
To run a single job (as a capsule does):
from biodata_cache.sync import run_sync_job
run_sync_job() # reads BIODATA_CACHE_SYNC_JOB, or pass e.g. run_sync_job("qc")
To rebuild everything in one local process (not used by the pipeline):
from biodata_cache.sync import update_all_tables
update_all_tables()
1 maintainer
Release files for biodata-cache 0.42.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| biodata_cache-0.42.0.tar.gz | 227.7 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| biodata_cache-0.42.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 476.2 kB
Release files / biodata_cache-0.42.0.tar.gz
| Download URL | biodata_cache-0.42.0.tar.gz |
|---|---|
| Size | 227.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
56e8c740338bd08546f1b6b89e06612e9d9458d2c7df726627ccb22bdbf3ccb9
|
|
BLAKE2b-256 checksum How to use checksums |
d6796520f38c223c79e6ae27275f5ad97430a5cd419586edeb5cb7f2b9556781
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Release files / biodata_cache-0.42.0-py3-none-any.whl
| Download URL | biodata_cache-0.42.0-py3-none-any.whl |
|---|---|
| Size | 248.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
eca188322625f7dea7962f6293d1bab46f26cabc92797e99d99d86ee1e54e9ce
|
|
BLAKE2b-256 checksum How to use checksums |
ae97d9a6573c41e165806e86a05ee524a405beb09806710fc1b9a018ecd8fef7
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
