Skip to content

Latest commit

 

History

History
366 lines (268 loc) · 14.2 KB

README.md

File metadata and controls

366 lines (268 loc) · 14.2 KB
Raw

ome2024-ngff-challenge

Actions Status PyPI version PyPI platforms Image.SC Zulip

Project planning and material repository for the 2024 challenge to generate 1 PB of OME-Zarr data.

UPDATE (20th October 2024): The data converted for the challenge can now be browsed and searched at https://ome.github.io/ome2024-ngff-challenge/.

Challenge overview

The high-level goal of the challenge is to generate OME-Zarr data according to a development version of the specification to drive forward the implementation work and establish a baseline for the conversion costs that members of the community can expect to incur.

Data generated within the challenge will have:

  • all v2 arrays converted to v3, optionally sharding the data
  • all .zattrs metadata migrated to zarr.json["attributes"]["ome"]
  • a top-level ro-crate-metadata.json file with minimal metadata (specimen and imaging modality)

You can example the contents of a sample dataset by using the minio client:

$ mc config host add uk1anon https://uk1s3.embassy.ebi.ac.uk "" ""
Added `uk1anon` successfully.
$ mc ls -r uk1anon/idr/share/ome2024-ngff-challenge/0.0.5/6001240.zarr/
[2024-08-01 14:24:35 CEST]  24MiB STANDARD 0/c/0/0/0/0
[2024-08-01 14:24:28 CEST]   598B STANDARD 0/zarr.json
[2024-08-01 14:24:32 CEST] 6.0MiB STANDARD 1/c/0/0/0/0
[2024-08-01 14:24:28 CEST]   598B STANDARD 1/zarr.json
[2024-08-01 14:24:29 CEST] 1.6MiB STANDARD 2/c/0/0/0/0
[2024-08-01 14:24:28 CEST]   592B STANDARD 2/zarr.json
[2024-08-01 14:24:28 CEST] 1.2KiB STANDARD ro-crate-metadata.json
[2024-08-01 14:24:28 CEST] 2.7KiB STANDARD zarr.json

Other samples:

Expand for more details on creation of these samples

4496763.json was created with ome2024-ngff-challenge commit 0e1809bf3b.

First the config details were generated with:

$ ome2024-ngff-challenge --input-bucket=idr --input-endpoint=https://uk1s3.embassy.ebi.ac.uk --input-anon zarr/v0.4/idr0047A/4496763.zarr params_4496763.json --output-write-details

The params_4496763.json file was edited to set "shards" to: [4, 1, sizeY, sizeX] for each pyramid resolution to create a single shard for each Z section.

# params_4496763.json
[{"shape": [4, 25, 2048, 2048], "chunks": [1, 1, 2048, 2048], "shards": [4, 1, 2048, 2048]}, {"shape": [4, 25, 1024, 1024], "chunks": [1, 1, 1024, 1024], "shards": [4, 1, 1024, 1024]}, {"shape": [4, 25, 512, 512], "chunks": [1, 1, 512, 512], "shards": [4, 1, 512, 512]}, {"shape": [4, 25, 256, 256], "chunks": [1, 1, 256, 256], "shards": [4, 1, 256, 256]}, {"shape": [4, 25, 128, 128], "chunks": [1, 1, 128, 128], "shards": [4, 1, 128, 128]}, {"shape": [4, 25, 64, 64], "chunks": [1, 1, 64, 64], "shards": [4, 1, 64, 64]}]

This was then used to run the conversion:

ome2024-ngff-challenge --input-bucket=idr --input-endpoint=https://uk1s3.embassy.ebi.ac.uk --input-anon zarr/v0.4/idr0047A/4496763.zarr 4496763.zarr --output-read-details params_4496763.json

9822152.zarr was created with ome2024-ngff-challenge commit f17a6de963.

The chunks and shard shapes are specified to be the same for all resolution levels. This is required since the smaller resolution levels of the source image at https://ome.github.io/ome-ngff-validator/?source=https://uk1s3.embassy.ebi.ac.uk/idr/zarr/v0.4/idr0083A/9822152.zarr have chunks that correspond to the resolution shape, e,g, 1,1,1,91,141 and this will fail to convert using a shard shape of 1,1,1,4096,4096.

Took 34 minutes to run conversion with this command:

$ ome2024-ngff-challenge --input-bucket=idr --input-endpoint=https://uk1s3.embassy.ebi.ac.uk --input-anon zarr/v0.4/idr0083A/9822152.zarr 9822152.zarr --output-shards=1,1,1,4096,4096 --output-chunks=1,1,1,1024,1024 --log debug

Took 9 hours to run this conversion (before multi-threading changes):

$ ome2024-ngff-challenge 9846151.zarr/0 will/9846151_2D_chunks_3.zarr --output-shards=1,1,1,4096,4096 --output-chunks=1,1,1,1024,1024 --log debug

Plate conversion, took 19 minutes, choosing a shard size that contained a whole image. Image shape is 1,3,1,1024,1280.

$ ome2024-ngff-challenge --input-bucket=bia-integrator-data --input-endpoint=https://uk1s3.embassy.ebi.ac.uk --input-anon S-BIAD847/0762bf96-4f01-454d-9b13-5c8438ea384f/0762bf96-4f01-454d-9b13-5c8438ea384f.zarr /data/will/idr0035/Week9_090907.zarr --output-shards=1,3,1,1024,2048 --output-chunks=1,1,1,1024,1024 --log debug

CLI Commands

resave: convert your data

The ome2024-ngff-challenge tool can be used to convert an OME-Zarr 0.4 dataset that is based on Zarr v2. The input data will not be modified in any way and a full copy of the data will be created at the chosen location.

Getting started

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr

is the most basic invocation of the tool. If you do not choose a license, the application will fail with:

No license set. Choose one of the Creative Commons license (e.g., `--cc-by`) or skip RO-Crate creation (`--rocrate-skip`)

Licenses

There are several license options to choose from. We suggest one of:

  • --cc-by: credit must be given to the creator
  • --cc0: Add your data to the public domain

Alternatively, you can choose your own license, e.g.,

--rocrate-license=https://creativecommons.org/licenses/by-nc/4.0/

to restrict commercial use of your data. Additionally, you can disable metadata collection at all.

Note: you will need to add metadata later for your dataset to be considered valid.

Metadata

There are four additional fields of metadata that are being collected for the challenge:

  • organism and modality: RECOMMENDED
  • name and description: SUGGESTED

These can be set via the properties prefixed with --rocrate- since they will be stored in the standard RO-Crate JSON file (./ro-crate-metadata.json) at the top-level of the Zarr dataset.

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --rocrate-organism=NCBI:txid9606      # Human
ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --rocrate-modality=obo:FBbi_00000369  # SPIM
ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --rocrate-name="short name of dataset"
ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --rocrate-description="and a longer description"

For other examples including several other NCBI and FBbi terms, please see:

ome2024-ngff-challenge resave --help

Re-running the script

If you would like to re-run the script with different parameters, you can additionally set --output-overwrite to ignore a previous conversion:

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --output-overwrite

Writing in parallel

By default, 16 chunks of data will be processed simultaneously in order to bound memory usage. You can increase this number based on your local resources:

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --output-threads=128

Reading/writing remotely

If you would like to avoid downloading and/or upload the Zarr datasets, you can set S3 parameters on the command-line which will then treat the input and/or output datasets as a prefix within an S3 bucket:

ome2024-ngff-challenge resave --cc-by \
        --input-bucket=BUCKET \
        --input-endpoint=HOST \
        --input-anon \
        input.zarr \
        output.zarr

A small example you can try yourself:

ome2024-ngff-challenge resave --cc-by \
        --input-bucket=idr \
        --input-endpoint=https://uk1s3.embassy.ebi.ac.uk \
        --input-anon \
        zarr/v0.4/idr0062A/6001240.zarr \
        /tmp/6001240.zarr

Reading/writing via a script

Another R/W option is to have resave.py generate a script which you can execute later. If you pass --output-script, then rather than generate the arrays immediately, a file named convert.sh will be created which can be executed later.

For example, running:

ome2024-ngff-challenge resave --cc-by dev2/input.zarr /tmp/scripts.zarr --output-script

produces a dataset with one zarr.json file and 3 convert.sh scripts:

/tmp/scripts.zarr/0/convert.sh
/tmp/scripts.zarr/1/convert.sh
/tmp/scripts.zarr/2/convert.sh

Each of the scripts contains a statement of the form:

zarrs_reencode --chunk-shape 1,1,275,271 --shard-shape 2,236,275,271 --dimension-names c,z,y,x --validate dev2/input.zarr /tmp/scripts.zarr

Running this script will require having installed zarrs_tools with:

cargo install zarrs_tools
export PATH=$PATH:$HOME/.cargo/bin

Optimizing chunks and shards

Zarr v3 supports shards, which are files that contain multiple chunks. The shape of a shard must be a multiple of the chunk size in every dimension. There is not yet a single heuristic for determining the chunk and shard sizes that will work for all data. The default shard shape chosen by resave is the full shape of the image array.

In order to limit the size of a shard, if the shard exceeds 100,000,000 pixels then you must specify the shard-shape. You can specify the shard shape, using --output-shards, which will be used for all pyramid resolutions. This may cause issues if the chunk shape changes for lower resolutions (to match the smaller image shape). In this case, you should also specify the chunk-shape to be used for all resolutions:

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --output-chunks=1,1,1,256,256 --output-shards=1,1,1,2048,2048

Alternatively, you can use a JSON file to review and manually optimize the chunking and sharding parameters on a per-resolution basis:

ome2024-ngff-challenge resave --cc-by input.zarr parameters.json --output-write-details

This will write a JSON file of the form:

[{"shape": [...], "chunks": [...], "shards": [...]}, ...

where the order of the dictionaries matches the order of the "datasets" field in the "multiscales". Edits to this file can be read back in using the output-read-details flag:

ome2024-ngff-challenge resave --cc-by input.zarr output.zarr --output-read-details=parameters.json

Note: Changes to the shape are ignored.

More information

See ome2024-ngff-challenge resave -h for more arguments and examples.

lookup: finding ontology terms (WIP)

The ome2024-ngff-challenge tool can also be used to look up terms from the EBI OLS for setting metadata fields like --rocrate-modality and --rocrate-organism:

ome2024-ngff-challenge lookup "homo sapiens"
ONTOLOGY  	TERM                	LABEL                         	DESCRIPTION
ncbitaxon 	NCBITaxon_9606      	Homo sapiens
vto       	VTO_0011993         	Homo sapiens
snomed    	SNOMED_337915000    	Homo sapiens
...

Related work

The following additional PRs are required to work with the data created by the scripts in this repository:

Slightly less related but important at the moment: