Configuring BATCHIE

Pipeline Wide Parameters

Pipeline-wide parameters affect the execution of the pipeline as a whole. These parameters can be specificed on the command line when running batchie.py.

--n_chunks is the number of chunks to split distance matrix/plate score calculation into. --n_chains is the number of parallel MCMC chains to run. --batch-size is the number of plates to select per iteration --outdir is the output directory --screen is the screen to use for running the pipeline

Process Specific Parameters

BATCHIE as a pipeline is made up of several component Python command line interface applications.

Documentation for each individual command line application is available here: Command Line Interface. Command line parameters can be passed into each application using a single nextflow .config file.

Several pre-defined configuration files are available in nextflow/config . We recommend using these as templates for your own configuration files.

Let’s take a look at an example configuration file to explain its contents:

process {
    publishDir = { "${params.outdir}" }

    withName: TRAIN_MODEL {
        ext.args = [
            '--model', 'SparseDrugCombo',
            '--model-param', 'n_embedding_dimensions=10',
            '--n-burnin', '10',
            '--n-samples', '10',
            '--thin', '1'
        ].join(' ')
    }

    withName: CALCULATE_DISTANCE_MATRIX_CHUNK {
        ext.args = [
            '--distance-metric', 'MSEDistance'
        ].join(' ')
    }

    withName: CALCULATE_SCORE_CHUNK {
        ext.args = [
            '--scorer', 'GaussianDBALScorer',
            '--seed', '12'
        ].join(' ')
    }
}

This is boilerplate that tells nextflow that the output for all processes should be published to the directory specified in the pipeline-wide --outdir parameter:

process {
    publishDir = { "${params.outdir}" }

This section:

withName: TRAIN_MODEL {
    ext.args = [
        '--model', 'SparseDrugCombo',
        '--model-param', 'n_embedding_dimensions=10',
        '--n-burnin', '10',
        '--n-samples', '10',
        '--thin', '1'
    ].join(' ')
}

Specifies what command line options will be passed to the call to train_model. You’ll define a similar section for each other application which you want to pass command line options to. Some applications dont have any required command line options, so you can leave those sections out entirely.

You can always refer to the Command Line Interface documentation for more information on what options are available for each process.