The goal of omnideconv is to unify second generation cell-type deconvolution methods in an R package.

There are two ways to install omnideconv:

• The minimal installation installs only the dependencies required for the basic functionalities. All deconvolution methods need to be installed on-demand.
• The complete installation installs all dependencies including all deconvolution methods. This may take a considerable time.

Since not all dependencies are on CRAN or Bioconductor, omnideconv is available from GitHub only. We recommend installing it through the pak package manager:

# install the `pak` package manager
install.packages("pak")

# minimal installation
pak::pkg_install("omnideconv/omnideconv")

# complete installation, including Python dependencies
pak::pkg_install("omnideconv/omnideconv", dependencies = TRUE)
omnideconv::install_all_python()

Upon the first loading, miniconda will be installed if not already present. A dedicated conda environment will be created to host the python-based methods.

The methods currently implemented in omnideconv are:

• AutoGeneS (“autogenes”)
• Bisque (“bisque”)
• BayesPrism ("bayesprism")
• BSeq-sc (“bseqsc”)
• CDSeq (“cdseq”)
• CIBERSORTx (“cibersortx”)
• CPM (“cpm”)
• DWLS (“dwls”)
• MOMF (“momf”)
• MuSiC (“music”)
• Scaden (“scaden”)
• SCDC (“scdc”)

All the deconvolution methods included in omnideconv can be run in one step, trough the function deconvolute, which takes in input the matrix of bulk RNAseq to be deconvolved (bulk_gene_expression), along with the training single cell expression matrix (single_cell_object) with the cell type annotations and sample information.

deconvolution <- omnideconv::deconvolute(bulk_gene_expression, method,
                                         single_cell_object, cell_type_annotations, batch_ids)

The methods AutoGeneS, BSeq-Sc, DWLS, CIBERSORTx, MOMF and Scaden first optimize their internal model, for example building a signature matrix, and then use this model to perform deconvolution. For these methods, the build_model function can be used. The obtained model can then be given in input to the deconvolute function, omitting the single cell data.

signature <- omnideconv::build_model(single_cell_object, cell_type_annotations,
                                     batch_ids, method, bulk_gene_expression)

deconvolution <- omnideconv::deconvolute(bulk_gene_expression, signature)

The deconvolute function returns a sample x cell type matrix with the estimated cell fractions

Different methods have different requirements in terms of input data. This list has been compiled considering the methods documentation, described data procssing or authors recommendation

For more information and an example workflow see the vignette of this package.

Most methods do not require additional software/tokens, but there are a few exceptions:

• A working version of Docker or Singularity is required for CIBERSORTx
• A token for CIBERSORTx is required from this website: https://cibersortx.stanford.edu/
• The CIBERSORT source code is required for BSeq-sc (see tutorial in ?omnideconv::bseqsc_config)

Note that, while omnideconv itself is free (GPL 3.0), you may need to obtain a license to use the individual methods. See the table below for more information. If you use this package in your work, please cite both our package and the method(s) you are using.

CITATION