workflow

Snakemake workflows

needs to be exectued from project directory (as root directory)

Single experiment

Run a single comparison

snakemake -n # uses workflow/Snakefile

which executes the following rules

Grid Search for Hyperparameters

# pwd: path/to/project
snakemake --snakefile workflow/Snakefile_grid.smk -n -p
# parallel execution on the same machine (shared GPU, CPU) does not work
snakemake --snakefile workflow/Snakefile_grid.smk --jobs 1

Configuration files are stored in project/config directory

provide one or more config files explicitly to overwrite defaults

snakemake --snakefile workflow/Snakefile_grid.smk --configfile config/other.yaml -n -p

e.g. for small N grid search:

snakemake --snakefile workflow/Snakefile_grid.smk --configfile config/grid_search_small_data/config_grid.yaml -p -c1 -n

rule graph is

Command: snakemake --snakefile workflow/Snakefile_grid.smk --rulegraph | dot -Tpng > rulegraph_gridsearch.png

Repeated training

Repeated training of same models on same dataset

• see if model trains stable on one dataset

snakemake --snakefile workflow/Snakefile_best_repeated_train.smk -n -p

Repeated training of models across machine datasets

• see how model perform across similar datasets

snakemake --snakefile workflow\Snakefile_best_across_datasets.smk -p -c1 -n

ALD data

This will need a two step procedure as a single experiment for the semi-supervised models needs to be run. Then the comparsion with the shifted normal distribution and the ald setup is done.

# fit models
snakemake --configfile config\appl_ald_data\plasma\proteinGroups\config.yaml -p -c1 -n
# compare downstream performance
snakemake --snakefile workflow\Snakefile_ald_comparison.smk --configfile config/appl_ald_data/plasma/proteinGroups/comparison.yaml -p -c1 -n

For the reduced dataset (80% of the original dataset), which was generated using 00_8_add_random_missing_values.py:

# fit models
snakemake --configfile config/appl_ald_data/plasma/proteinGroups_80%_dataset/config.yaml -p -c1
# compare downstream performance
snakemake --snakefile workflow\Snakefile_ald_comparison.smk --configfile config/appl_ald_data/plasma/proteinGroups_80%_dataset/comparison.yaml -p -c1 -n

Rule graph for ALD comparison (after using single experiment)

Repeated execution

Executes both workflows for model training and comparison ten times:

• F to force execution of all workflows (each workflow then checks what it needs to execute)
• nolock : ensure that parent workflow does not block child workflow

snakemake -s workflow\Snakefile_ald_comparison_repeated.smk -p -c1 --nolock --drop-metadata -F -n

Test misc_ notebook

TestNotebooks.smk test the misc_* notebooks in the project folder.