Skip to content

image_analysis_README.md

Latest commit

 

History

History
72 lines (51 loc) · 3.56 KB

image_analysis_README.md

File metadata and controls

72 lines (51 loc) · 3.56 KB

Exercise 2: Image Analysis

In this exercise, you will create a workflow for the BBBC010 dataset that

  1. for each worm in each image, extract shape, intensity, and texture features from the brighfield channel, and generate a class label from the intensity of the GFP channel ("live" or "dead", see the BBBC010 website for more details);

  2. combine all the features and labels across images into an aggregated data structure;

  3. train a decision tree classifier to distinguish "live" from "dead" worms based on the brightfield channel features;

  4. generate a visual summary of the classification results.

WARNING: you must have successfully run the data preparation workflow from exercise 1 (data_preparation_workflow.nf) to be able to work on this exercise!

If you did not manage to complete exercise 1, you can generate the required outputs as follows:

  cd ../../solutions/01-data-preparation/
  nextflow run data_preparation_workflow.nf

Learning goals

  1. Multi-output process handling

    • Manage processes that generate multiple output files

  2. Data aggregation patterns

    • Collect outputs from parallel processes with .collect()

  3. Complex workflows

    • Build analysis pipelines composed of multiple steps
    • Pass multiple files between processes
    • Manage global parameters throughout the workflow

How to go about this exercise

  • Open image_analysis_workflow.nf with your favourite code editor and complete the sections marked with YOUR TURN! Replace the "..." below with your code! - you may have to open an examine the provided Python scripts when prompted to do so!

  • Once you have completed the missing pieces of code, run your workflow:

     nextflow run image_analysis_workflow.nf

  • Debug when needed by

    • inspecting the nextflow log: NextFlow saves an execution log in a .nextflow.log file that can be opened in a text editor

    • checking the current directory for failed tasks: NextFlow creates a work subdirectory into which it saves a trace of everything it does! In the error traceback that appears in your Terminal, look for the working directory where the error happened, e.g.:

    Work dir:
/Users/myname/my/path/to/nextflow101/exercises/2-image-analysis/work/bf/d7f0c2e2609...

    Using the actual path from the error message, you can then navigate to the corresponding working directory:

    cd work/bf/d7f0c2e2609...

    There, you can examine what happened by looking at the exact command that NextFlow executed (.command.sh) and the error output (.command.err), which can both be opened in a text editor

    • re-running the workflow with a more verbose output:
    nextflow run main.nf -with-trace

  • When your workflow completes successfully, check the outputs of your workflow in the ../../data/ folder

  • If you are stuck, check the solution in ../../solutions/02-image-analysis/!

  • BONUS: experiment with different parameter values for intensity_threshold and train_split, and observe how it affects the results!

Further resources