Add a control method

This guide will show you how to create a new control method.

A control method is used to test the relative performance of all other methods, and also as a quality control for the pipeline as a whole. A control method can either be a positive control or a negative control. The positive control and negative control methods set a maximum and minimum threshold for performance, so any new method should perform better than the negative control methods and worse than the positive control method.

This guide will show you how to create a new Viash component. In the following we will show examples for both Python and R. Note that the Task template repo is used throughout the guide, so make sure to replace any occurrences of "task_template" with your task of interest.

Tip

Make sure you have followed the “Getting started” guide.

Step 1: Create a new component

Use the create_*_method.sh script found in the scripts repository to start creating a new control method. Open the script and update the name parameter to the desired name of the method and update the type to control_method.

scripts/create_component/create_python_method.sh
scripts/create_component/create_python_method.sh
common/scripts/create_component \
  --name my_python_method \
  --language python \
  --type control_method
Check inputs
Check language
Check API file
Read API file
Create output dir
Create config
Create script
Done!

This creates a new folder at src/control_methods/my_python_method containing a Viash config and a script.

tree src/tasks/label_projection/control_methods/my_python_method
    ├── script.py         Script for running the method.
    ├── config.vsh.yaml   Config file for method.
    └── ...               Optional additional resources.
scripts/create_component/create_r_method.sh
Check inputs
Check language
Check API file
Read API file
Create output dir
Create config
Create script
Done!
scripts/create_component/create_r_method.sh
common/scripts/create_component \
  --name my_r_method \
  --language r \
  --type control_method
Check inputs
Check language
Check API file
Read API file
Create output dir
Create config
Create script
Done!

This creates a new folder at src/control_methods/my_r_method containing a Viash config and a script.

tree src/tasks/label_projection/control_methods/my_r_method
    ├── script.R          Script for running the method.
    ├── config.vsh.yaml   Config file for method.
    └── ...               Optional additional resources.

Change the --name to a unique name for your metric. It must match the regex [a-z][a-z0-9_]* (snakecase).

Change the --type to control_method.

  • A config file contains metadata of the component and the dependencies required to run it. In steps 2 and 3 we will fill in the required information.
  • A script contains the code to run the method. In step 4 we will edit the script.
Tip

Some tasks have multiple method subtypes (e.g. batch_integration), which will require you to use a different value for --type corresponding to the desired method subtype.

Step 2: Fill in metadata

The Viash config contains metadata of your method, which script is used to run it, and the required dependencies.

Generated config file

This is what the config.vsh.yaml generated by the create_component component looks like:

Contents of config.vsh.yaml
# The API specifies which type of component this is.
# It contains specifications for:
#   - The input/output files
#   - Common parameters
#   - A unit test
__merge__: ../../api/comp_control_method.yaml


# A unique identifier for your component (required).
# Can contain only lowercase letters or underscores.
name: my_python_method

# Metadata for your component
info:
  # A relatively short label, used when rendering visualisations (required)
  label: My Python Method
  # A one sentence summary of how this method works (required). Used when 
  # rendering summary tables.
  summary: "FILL IN: A one sentence summary of this method."
  # A multi-line description of how this component works (required). Used
  # when rendering reference documentation.
  description: |
    FILL IN: A (multi-line) description of how this method works.
  # Which normalisation method this component prefers to use (required).
  preferred_normalization: log_cp10k

# Component-specific parameters (optional)
# arguments:
#   - name: "--n_neighbors"
#     type: "integer"
#     default: 5
#     description: Number of neighbors to use.

# Resources required to run the component
resources:
  # The script of your component (required)
  - type: python_script
    path: script.py
  # Additional resources your script needs (optional)
  # - type: file
  #   path: weights.pt

engines:
  # Specifications for the Docker image for this component.
  - type: docker
    image: ghcr.io/openproblems-bio/base_images/python:1.1.0
    # Add custom dependencies here (optional). For more information, see
    # https://viash.io/reference/config/engines/docker/#setup .
    # setup:
    #   - type: python
    #     packages: scib==1.1.5

runners:
  # This platform allows running the component natively
  - type: executable
  # Allows turning the component into a Nextflow module / pipeline.
  - type: nextflow
    directives:
      label: [midtime,midmem,midcpu]
Contents of config.vsh.yaml
# The API specifies which type of component this is.
# It contains specifications for:
#   - The input/output files
#   - Common parameters
#   - A unit test
__merge__: ../../api/comp_control_method.yaml


# A unique identifier for your component (required).
# Can contain only lowercase letters or underscores.
name: my_r_method

# Metadata for your component
info:
  # A relatively short label, used when rendering visualisations (required)
  label: My R Method
  # A one sentence summary of how this method works (required). Used when 
  # rendering summary tables.
  summary: "FILL IN: A one sentence summary of this method."
  # A multi-line description of how this component works (required). Used
  # when rendering reference documentation.
  description: |
    FILL IN: A (multi-line) description of how this method works.
  # Which normalisation method this component prefers to use (required).
  preferred_normalization: log_cp10k

# Component-specific parameters (optional)
# arguments:
#   - name: "--n_neighbors"
#     type: "integer"
#     default: 5
#     description: Number of neighbors to use.

# Resources required to run the component
resources:
  # The script of your component (required)
  - type: r_script
    path: script.R
  # Additional resources your script needs (optional)
  # - type: file
  #   path: weights.pt

engines:
  # Specifications for the Docker image for this component.
  - type: docker
    image: ghcr.io/openproblems-bio/base_images/r:1.1.0
    # Add custom dependencies here (optional). For more information, see
    # https://viash.io/reference/config/engines/docker/#setup .
    # setup:
    #   - type: r
    #     packages: tidyverse

runners:
  # This platform allows running the component natively
  - type: executable
  # Allows turning the component into a Nextflow module / pipeline.
  - type: nextflow
    directives:
      label: [midtime,midmem,midcpu]

Required metadata fields

Please edit info section in the config file to fill in the necessary metadata.

  • .__merge__: The API specifies which type of component this is. It contains specifications for:

    • The input/output files
    • Common parameters
    • A unit test
  • .name: A unique identifier. Can only contain lowercase letters, numbers or underscores.

  • .label: A unique, human-readable, short label. Used for creating summary tables and visualisations.

  • .summary: A one sentence summary of purpose and methodology. Used for creating an overview tables.

  • .description: A longer description (one or more paragraphs). Used for creating reference documentation and supplementary information.

Step 3: Add dependencies

Each component has it’s own set of dependencies, because different components might have conflicting dependencies.

base images

For your convenience we have created several base images that can be used for python or R scripts. These images can be found in the OpenProblems docker repo base_images. Click on the packages to view the url you need to use. You are not required to use these images but install the required packages to make sure OpenProblems works properly.

  • openproblems/base_python Base image for python scripts.

  • openproblems/base_r Base image for R scripts.

  • openproblems/base_pytorch_nvidia Base image for scripts that use pytorch with nvidia gpu support.

  • openproblems/base_tensorflow_nvidia Base image for scripts that use tensorflow with nvidia gpu support.

custom image

Update the setup definition in the platforms section of the config file. This section describes the packages that need to be installed in the Docker image and are required for your method to run.

If you’re using a custom image use the following minimum setup:

platforms:
  - type: docker
    Image: your custom image
    setup:
      - type: apt
        packages:
          - procps
      - type: python
        packages:
          - anndata~=0.10.0
          - scanpy~=1.10.0
          - pyyaml
          - requests
          - jsonschema
        github: openproblems-bio/core#subdirectory=packages/python/openproblems
platforms:
  - type: docker
    Image: your custom image
    setup:
      - type: apt
        packages:
          - procps
          - libhdf5-dev
          - libgeos-dev
          - python3
          - python3-pip
          - python3-dev
          - python-is-python3
      - type: python
        packages:
          - rpy2
          - anndata~=0.10.0
          - scanpy~=1.10.0
          - pyyaml
          - requests
          - jsonschema
          github: openproblems-bio/core#subdirectory=packages/python/openproblems
      - type: r
        packages:
          - anndata
          - BiocManager
          - reticulate
          - bit64
        github:
          - openproblems-bio/core/packages/r/openproblems

Please check out this guide for more information on how to add extra package dependencies.

Note

Tip: After making changes to the components dependencies, you will need to rebuild the docker container as follows:

viash run src/control_methods/my_python_method/config.vsh.yaml -- \
  ---setup cachedbuild
[notice] Building container 'ghcr.io/openproblems-bio/task_template/control_methods/my_python_method:dev' with Dockerfile
output
[notice] Building container 'ghcr.io/openproblems-bio/task_template/control_methods/my_python_method:dev' with Dockerfile

Step 4: Edit script

A component’s script typically has five sections:

  1. Imports and libraries
  2. Argument values
  3. Read input data
  4. Generate results
  5. Write output data to file

Generated script

This is what the script generated by the create_component component looks like:

Contents of script.py
import anndata as ad

## VIASH START
# Note: this section is auto-generated by viash at runtime. To edit it, make changes
# in config.vsh.yaml and then run `viash config inject config.vsh.yaml`.
par = {
  'input_train': 'resources_test/.../train.h5ad',
  'input_test': 'resources_test/.../test.h5ad',
  'input_solution': 'resources_test/.../solution.h5ad',
  'output': 'output.h5ad'
}
meta = {
  'name': 'my_python_method'
}
## VIASH END

print('Reading input files', flush=True)
input_train = ad.read_h5ad(par['input_train'])
input_test = ad.read_h5ad(par['input_test'])
input_solution = ad.read_h5ad(par['input_solution'])

print('Preprocess data', flush=True)
# ... preprocessing ...

print('Train model', flush=True)
# ... train model ...

print('Generate predictions', flush=True)
# ... generate predictions ...

print("Write output AnnData to file", flush=True)
output = ad.AnnData(
  
)
output.write_h5ad(par['output'], compression='gzip')
Contents of script.R
library(anndata)

## VIASH START
par <- list(
  input_train = "resources_test/.../train.h5ad",
  input_test = "resources_test/.../test.h5ad",
  input_solution = "resources_test/.../solution.h5ad",
  output = "output.h5ad"
)
meta <- list(
  name = "my_r_method"
)
## VIASH END

cat("Reading input files\n")
input_train <- anndata::read_h5ad(par[["input_train"]])
input_test <- anndata::read_h5ad(par[["input_test"]])
input_solution <- anndata::read_h5ad(par[["input_solution"]])

cat("Preprocess data\n")
# ... preprocessing ...

cat("Train model\n")
# ... train model ...

cat("Generate predictions\n")
# ... generate predictions ...

cat("Write output AnnData to file\n")
output <- anndata::AnnData(
  
)
output$write_h5ad(par[["output"]], compression = "gzip")

The required sections are explained here in more detail:

a. Imports and libraries

In the top section of the script you can define which packages/libraries the method needs. If you add a new or different package add the dependency to config.vsh.yaml in the setup field (see above).

b. Argument block

The Viash code block is designed to facilitate prototyping, by enabling you to execute directly by running python script.py (or Rscript script.R for R users). Note that anything between “VIASH START” and “VIASH END” will be removed and replaced with a CLI argument parser when the components are being built by Viash.

Here, the par dictionary contains all the arguments defined in the config.vsh.yaml file (including those from the defined __merge__ file). When adding a argument in the par dict also add it to the config.vsh.yaml in the arguments section.

c. Read input data

This section reads any input AnnData files passed to the component.

d. Generate results

This is the most important section of your script, as it defines the core functionality provided by the component. It processes the input data to create results for the particular task at hand.

e. Write output data to file

The output stored in a AnnData object and then written to an .h5ad file. The format is specified by the API file specified in the __merge__ field in the config file.

Step 5: Add resources (optional)

It is possible to add additional resources such as a file containing helper functions or other resources. Please visit this page for more information on how to do this.

Step 6: Try component

Your component’s API file contains the necessary unit tests to check whether your component works and the output is in the correct format.

You can test your component by using the following command:

viash test src/tasks/label_projection/control_methods/my_python_method/config.vsh.yaml
Output
Running tests in temporary directory: '/tmp/viash_test_true_labels_11901889007781665836'
====================================================================
+/tmp/viash_test_true_labels_11901889007781665836/build_engine_environment/true_labels ---verbosity 6 ---setup cachedbuild ---engine docker
[notice] Building container 'ghcr.io/openproblems-bio/task_template/control_methods/true_labels:test' with Dockerfile
[info] docker build -t 'ghcr.io/openproblems-bio/task_template/control_methods/true_labels:test'  '/tmp/viash_test_true_labels_11901889007781665836/build_engine_environment' -f '/tmp/viash_test_true_labels_11901889007781665836/build_engine_environment/tmp/dockerbuild-true_labels-MHiAVH/Dockerfile'
#0 building with "default" instance using docker driver

#1 [internal] load build definition from Dockerfile
#1 transferring dockerfile: 486B done
#1 DONE 0.0s

#2 [internal] load metadata for docker.io/openproblems/base_python:1.0.0
#2 DONE 0.1s

#3 [internal] load .dockerignore
#3 transferring context: 2B done
#3 DONE 0.0s

#4 [1/1] FROM docker.io/openproblems/base_python:1.0.0@sha256:bbb0a093e275498bf905237c8cd26124d436f5d35d0f8dd1749c06d0a0a2e88c
#4 CACHED

#5 exporting to image
#5 exporting layers done
#5 writing image sha256:a38fad88e9504251af5e89f54682bc4c43369b029a6d6fa699dfc33050136288 done
#5 naming to ghcr.io/openproblems-bio/task_template/control_methods/true_labels:test done
#5 DONE 0.0s
====================================================================
+/tmp/viash_test_true_labels_11901889007781665836/test_run_and_check_output/test_executable
>> Running test 'run'
>> Checking whether input files exist
>> Running script as test
Reading input files
Preprocess data
Train model
Generate predictions
Write output AnnData to file
>> Checking whether output file exists
>> Reading h5ad files and checking formats
Reading and checking output
  AnnData object with n_obs × n_vars = 213 × 0
    obs: 'label_pred'
    uns: 'dataset_id', 'method_id', 'normalization_id'
All checks succeeded!
====================================================================
+/tmp/viash_test_true_labels_11901889007781665836/test_check_config/test_executable
Load config data
Check .namespace
Check .info.type
Check component metadata
Checking contents of .info.preferred_normalization
Check Nextflow runner
All checks succeeded!
====================================================================
SUCCESS! All 2 out of 2 test scripts succeeded!
Cleaning up temporary directory

Visit “Run tests” for more information on running unit tests and how to interpret common error messages.

You can also run your component on local files using the viash run command. For example:

viash run src/tasks/label_projection/control_methods/my_python_method/config.vsh.yaml -- \
  --input_train resources_test/task_template/pancreas/train.h5ad \
  --input_test resources_test/task_template/pancreas/test.h5ad \
  --input_solution resources_test/task_template/pancreas/solution.h5ad \
  --output output.h5ad

Next steps

If your component works, please create a pull request.