create unit tests

[kelly-sovacool]

• create a tiny test dataset, or reuse one from another scRNA package
• write basic unit tests for main user-facing functions
  • at minimum: make sure the function runs without error
  • preferably: also test edge cases, make sure expected errors occur when they should, input validation, etc.
• aim for > 80% code coverage

[bianjh-cloud]

Functions that might require unit testing:

cluster_metrics:

• check silhouette columns are included / omitted in each use case
• clusters <- as.numeric(unlist(so[[cluster_label]])) works for columns containing numeric, character, factors.

filter_doublets

• verify that each case of doublet_finder_method (DoubletFinder, scDblFinder, union, consensus) works
• subsetting works as intended (returned object contains the expected cell names, cell number)

make_bubble_plot

• function would give an empty dataframe / image if no features was found
• columns group, Gene, AvgExp, and PctExp are consistently output
• ggplot is returned and matches expected output for different cases (e.g. empty when no features are found)

make_volcano_plot

• similar to make_bubble_plot

preprocess_sample

• check that the right dataframe is returned every time with the test dataset

run_AUCell

run_batch_correction
might require set.seed() to ensure that batch corrected dimension reduction plots are returned consistently

run_singleR_db

• appropriate singleR columns returned with selected species
• potentially consider automatically recognizing species by seurat's gene names (all captial -> human, title case -> mouse)

split_featurePlot

• each plot's title correspond to ident value
• function errors clearing if so[[split_ident]] is missing
• returns a list of ggplots, keyed by features found

[kelly-sovacool]

@bianjh-cloud your checklist is a good starting point, and a good way to think about how the unit tests should be written to ensure assumptions hold true. I would add that all user-facing functions will need at least one unit test to ensure the essential functions are covered.

Here's the current list of all user-facing functions:

• add_VDJ_metadata
• calc_sc_gsea_score
• cluster_metrics
• convert_human_gene_list
• convert_mtx_to_h5
• filter_doublets
• make_bubble_plot
• make_volcano_plot
• preprocess_sample
• run_AUCell
• run_batch_correction
• run_hypergeometric_test
• run_singleR
• run_singleR_cluster
• run_singleR_db
• seurat_clustering
• split_featurePlot

Extracted with:

grep export NAMESPACE | grep -v "[>:=\.]" | sed -E 's/export\(([a-zA-Z0-9_]*)\)/- [ ] \1/g'

[bianjh-cloud]

@bianjh-cloud your checklist is a good starting point, and a good way to think about how the unit tests should be written to ensure assumptions hold true. I would add that all user-facing functions will need at least one unit test to ensure the essential functions are covered.

Here's the current list of all user-facing functions:

• add_VDJ_metadata[ ] calc_sc_gsea_score[ ] cluster_metrics[ ] convert_human_gene_list[ ] convert_mtx_to_h5[ ] filter_doublets[ ] make_bubble_plot[ ] make_volcano_plot[ ] preprocess_sample[ ] run_AUCell[ ] run_batch_correction[ ] run_hypergeometric_test[ ] run_singleR[ ] run_singleR_cluster[ ] run_singleR_db[ ] seurat_clustering[ ] split_featurePlot

Extracted with:

grep export NAMESPACE | grep -v "[>:=.]" | sed -E 's/export(([a-zA-Z0-9_]*))/- [ ] \1/g'

Thanks Kelly, will get started on this.

[kelly-sovacool]

We set up codecov for generating Code Coverage reports from the tests. Log in here with your github account to view the reports: https://app.codecov.io/gh/CCBR/SCOT

[bianjh-cloud]

Uploaded BRCA_Combine_and_Renormalize_SO_downsample.rds (20mb) test dataset to fixtures.

https://www.nature.com/articles/s41588-021-00911-1