CELLEX-GSEA

R package to calculate cell specific expression of genes, using an input gene set and a CELLEX dataset.

Installation instructions

Use the function install_github from the package devtools to install this R package (for the code to function properly, please update any other outdated packages in the process when/if prompted to):

devtools::install_github("https://github.com/Mporse/cellex-gsea")

Short tutorial

There are several functions in the package, however you should only use one for the analysis (all the other functions are helper functions to make the tool work correctly):

gsea_analysis()

Follow these to run an analysis of a input gene set using the function:

1. Load the package using:

library(cellex.gsea)

2. Save your gene set (in ENSEMBL or HGNC format) or protein set (in UNIPROT format) as a variable. A gene set that is enriched in the liver (obtained from the Human Protein Atlas, proteinatlas.org) is already included in the package, and can be loaded with the following command:

gene_set <- readRDS(system.file("gene_set_data", "liver.rds", package = "cellex.gsea"))

3. Run the enrichment analysis. If the input is in ENSEMBL format, you only need to supply the input gene set, as such:

gsea_analysis(gene_set)

That's it!

Important function arguments

Using other gene set input formats

The input gene set is not always in ENSEMBL format. Therefore, the argument input_type in the function needs to be changed:

• If the input set is in UNIPROT format:

gsea_analysis(gene_set, input_type = "uniprot")

• If the input set is in HGNC format:

gsea_analysis(gene_set, input_type = "hgnc")

Using other CELLEX datasets

The used CELLEX dataset can be provided by the user (so setting the argument cellex_data equal to the path to the dataset), or one of three datasets included in the package. Set the argument cellex_data equal to 1 to use the Tabula Muris CELLEX dataset, set equal to 2 to use the GTEX_v8 CELLEX dataset, or set sequal to 3 to use the HCL CELLEX dataset.

Specifying the preferred output

It can be specified whether or not to calulate empirical p-values:

• Set argument emp_p_value equal to TRUE to calculate empirical p-values for each cell type (this can take some time), otherwise set equal to FALSE (default).

NB: If calculating empirical P-values, I recommend using the Tabula Muris CELLEX dataset. Pre-computed null distributions have been made available online only for this dataset, and the package will download the proper null distribution (according to the size of the gene set). If calculating empirical P-values without using the pre-computed null distributions, make sure to increase the number of utilized cores through the n_cores argument (only possible on Unix systems), otherwise the computation can take a long time.

Specifying the test statistics

By default, the p-values are calculated using the Wilcoxon test. This can be changed by changing the value of the argument statistic:

• statistic = "KS" to use the Kolmogorov–Smirnov test.
• statistic = "T" to use the Welch's T-test.
• statistic = "W" to use the Mann-Whitney U (Wilcoxon rank-sum) test.
• statistic = "ES" to use the ssGSEA approach from the GSVA package to calculate enrichment scores (ES).

Other arguments

To see which other arguments are available and which values they accept, check the R documentation for the function:

?gsea_analysis

Web application

The tool has also been made available as a Shiny web application. It is being hosted on shinyapps.io using the free tier, which means the app is limited to 25 active hours per month. Thus, the use of the web application is quite limited and I would recommend using the R package version instead whenever possible, since it yields the same results.

• Shiny web application: https://porse.shinyapps.io/cellex-gsea/

Credits

The tool makes use of ESµ values from the CELLEX tool:

• GitHub: https://github.com/perslab/CELLEX
• Journal article (P. N. Timshel, J. J. Thompson, and T. H. Pers): https://elifesciences.org/articles/55851

The tool also makes use of code from the GSVA package for performing the ssGSEA approach:

• GitHub: https://github.com/rcastelo/GSVA
• Bioconductor: https://www.bioconductor.org/packages/release/bioc/html/GSVA.html
• Journal article (S. Hänzelmann, R. Castelo, and J. Guinney): https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-14-7