Cell Browser best practices
Best Practices
Formatting configuration files
- Typically you should keep a maximum of 80-120 characters per line; you can use
gqgqin VIM in visual mode to auto format a paragraph into multiple ~80 character lines - For special characters, please refer to HTML character encoding: https://ascii.cl/htmlcodes.htm
cellbrowser.conf
Putting things into cellbrowser-confs repo
From inside a dataset directory:
git add desc.conf cellbrowser.conf git commit -m “message” git push
Only do this for public datasets. If this is a collection, commit the files for each dataset in the collection. For additional help you can refer to Commit cellbrowser/desc.conf files.
Naming datasets
Dataset names should be:
- all lowercase
- 4 words or less
- less than 20 characters and separated by hyphens
The names need to be lowercase because the Cell Browser website code converts all names lowercase. There are only a few exceptions for early datasets (e.g. adultPancreas).
Layout Coordinates
- Capitalize
"UMAP"and"tSNE". - Remove extra layout coordinates (e.g. PCA or Harmony) because the cbImportTools export all of the possible layouts and they export only the only the first two coordinates. The CB can only handle two coordinates and so these layouts often look like a clump of cells.
The following two images are examples of PCA plots. For reference, the first image is from the "lung-airway" dataset and the second image is from the "hoc" dataset.
Finding a paper associated with a bioRxiv pub
Sometimes you will have to go back and edit the paper citation for a dataset.
1. Get the bioRxiv URL for your dataset, e.g. https://www.biorxiv.org/content/10.1101/2020.06.30.174391v1
2. Copy this bit of the URL: 10.1101/2020.06.30.174391
3. And feed it to the bioRxiv API: https://api.biorxiv.org/details/biorxiv/. Here's an example command to call the bioRxiv api via curl and then use jq to just get the key 'published' from the JSON response:
curl https://api.biorxiv.org/details/biorxiv/10.1101/2020.06.30.174391 2>/dev/null | jq '.[] | .[] | .published'
4. If it's published, it'll have a DOI (e.g. 10.1038/s41593-021-00872-y); otherwise it'll just say 'NA'.
5. Paste the DOI into https://www.doi.org/ and you'll be taken directly to the paper.
Finding all datasets with bioRxiv pubs
If you want to go through and find all datasets with bioRxiv pubs and update them:
cd /hive/data/inside/cells/datasets find . -name desc.conf | xargs grep "biorxiv" |grep -v "Strange\|\#"
Which should get you results like:
./cbl-dev/desc.conf:biorxiv_url = "https://www.biorxiv.org/content/10.1101/2020.06.30.174391v1 Aldinger et al. 2020. bioRxiv."
You can then extract the necessary pieces from the URLs and use them with the steps above.
Providing the Unit for datasets
unit=""
Provide the unit of the values used in the expression matrix. Typical values: "read count/UMI", "log of read count/UMI", "TPM", "log of TPM", "CPM", "FPKM", "RPKM".
For Seurat objects, the 'counts' slot is typically 'UMI count' for 10x data or 'read count' for Smart-seq2 or similar assays. The 'data' slot is the log-normalized version of the counts slot. This Github issue has some details: https://github.com/satijalab/seurat/issues/3711. For SCT assay datasets, it's slightly different: https://satijalab.org/seurat/reference/sctransform; in short though, the units are: counts -> (corrected) counts, data -> log1p(counts), scale.data -> pearson residuals.
It's probably easiest to ask the authors if you're unsure.
desc.conf
Most commonly used desc.conf settings to keep consistent:
title = "First word is capitalized and the rest is all lowercased"paper_url = “http://www.paper_url.com/xxx Last name et al. Journal. Year.”biorxiv_url = "https://www.biorxiv.org/content/123/123.full Last name et al. bioRxiv. Year."
The additional database links (GSE, Bioproject, SRA accessions, PMID, etc.) can be set to just the number/accession, no author info:
pmid = "12343234" geo_series = "GSE25097" dbgap = "phs000424.v7.p2" arrayexpress = "xxx" sra_study = "xxxx"
