swarchal / cptools2 Goto Github PK

Rather than forcing users to name their channels W1 ..Wn in the cellprofiler pipeline.

Not sure if this can be obtained from the cellprofiler pipeline file?

If a plate has been imaged twice there's two plate sub-directories within the plate directory which causes the LoadData dataframe to contain twice the number of expected rows.

We need a better way of checking for plate duplications, and for the LoadData module to handle this.

Due to some unique IT decisions, it's probably easy to include the scissorhands in this repo rather than have it as a dependency.

The CV8000 has the option of imaging some channels with multiple z-stacks and others without.
It's not likely this is going to happen with a screen but it's causing problems in testing, where the LoadData csv file has rows with empty channels for those with no image in that z-location.

Options
- Only image with equal z-stacks per channel, check for this and raise an informative error if not true
- Duplicate image paths for the missing channels for the z-locations we have

• staging is not required, but will it improve performance to operate on the scratch disks?
• definitely single core as don't need to worry about memory limits
• use singularity images rather than conda envs

config.yml -> JSON which contains all the PlateStore, and LoadData?

Might require re-writing a lot of stuff to only operate on dictionaries, the PlateStore is already a dictionary, not sure how best to serialize the dataframes.

At the moment it's just analysis_{random_hex}

This groups images by well and site before splitting, but relies on the ImageXpress specific parserix parser.

Options:

• Make replicate functions for the different microscopes
• Find a different way to avoid splitting images within a well+site

If we add two experiments that share plate names, I'm not currently sure what happens.

There are at least 2 issues that will happen with this:

1. LoadData csv files with identical names will overwrite each other, or they are possibly overwritten as dictionary keys in the plateStore implementation.
2. Trying to remove plates without specifying which experiment is ambiguous.

We need to define the behaviour, or at least raise a warning/error if this happens.

# experiment with plates named plate_A01, plate_A02... etc
experiment: /path/to/experiment_A

add plate:
    - experiment: /path/to/experiment_B
    - plates:
        - plate_A01
        - plate_A02
        - plate_A03

remove plate:
    - plates:
        - plate_A01

Options:

1. Detect if this happens and error, experiments should not share a config file if they have identically named plates.
2. Name LoadData csv files including the experiment name so they are not overwritten. In addition, remove plate in config needs to specify an experiment in addition to plate names.

There's some instances where it's checking for environment variables to maek sure it's on the right node which won't apply in this case.