Minimal Nextflow OME-Zarr workflow

Simple example created by a group at the "Next generation bioimage analysis workflows hackathon".

Usage

1. Git pull this repository
2. cd into the repository folder
3. Modify the data/input_params.yaml file to point to the absolute input image path
4. Run the following command: If you have a conda environment with nextflow installed, you can run the following command:

pixi run nextflow run . -params-file data/input_params_local.yaml -profile conda

or run with docker:

pixi run nextflow run . -params-file data/input_params_local.yaml -profile docker

Aims

• Explore what nf-core gives us for specifying inputs and outputs

• Explore storing versioning file as in nf-core. Update: putting all versioning logging file at the root for now.

• Create a github-repo for the below code

• Create a minimal workflow in Nextflow that uses OME-Zarr

  • Process 1: Create new gaussian blurred ome-zarr image
  • Process 2: Segment image
  • Process 3: Measure segment shape features

• Should the input be only one scale? Or multiple?

• How to handle the multi-scales for the outputs?

• Root ome-zarr + subfolder strings as input. Numpy/dask object as Image/labels data.

• label image stored within the same ome-zarr file as an image

• 'Hacked' nextflow IO to allow for reading/writing valid OME-Zarr files

• Where/how to store the table?

• A more tightly connected image visualisation tool?

• Integration Fractal tasks into Nextflow

• Bonus

  • Process only a part of an image
  • Use a Fractal task as one of the Nextflow processes

Observations

• Important tools and libraries require arrays of specific input dimensionality to operate on, thus we need convenient APIs and implementations that allow us to subset OME-Zarr.
  • Very few (none?) of the current tools natively work on multi-resolution input, thus be able to specify which resolution level to work on is important and not well supported by the python libraries that we found.

Analysis tools landscape in terms of input dimensionalities

2D RGB

• YOLO
• SAM

2D or 3D RGB

• Cellpose
  • Is it really RGB?

2D or 3D multi-channel

• elastix

ND

• skimage ?!
• ImgLib2 ?!