In saezlab/liana: LIANA: a LIgand-receptor ANalysis frAmework

knitr::opts_chunk$set(echo = TRUE)

LIANA's OmniPath resource was largely compiled from CCC databases with a human focus, as such LIANA works with human gene symbols by default.

Nevertheless, one could run LIANA with any other organism, provided a dictionary is available to map human genesymbols to those of species X.

In this vignette, we provide an example how this could be achieved using the HomoloGene database via OmnipathR with LIANA's generate_homologs function.

Note that LIANA provides a mouse Consensus (MouseConsensus) resource, using the homology-conversion approach described here.

Loading prerequisites

library(tidyverse)
library(OmnipathR)
library(liana)
library(magrittr)

liana_path <- system.file(package = "liana")
testdata <-
    readRDS(file.path(liana_path, "testdata", "input", "testdata.rds"))

# Convert testdata to putative mouse symbols
# Note that we explicitly provide this tutorial to avoid any such conversions when working with real data
# We simply do this here to provide an example
rownames(testdata@assays$RNA@counts) <- stringr::str_to_title(rownames(testdata@assays$RNA@counts))
rownames(testdata@assays$RNA@data) <- stringr::str_to_title(rownames(testdata@assays$RNA@data))

It is also recommended to ensure that you have the latest OmnipathR version installed:

remotes::install_github("saezlab/OmnipathR")

Internally available organisms

LIANA's generate_homologs function uses OmnipathR's homologene_download function to obtain a human to x species dictionary.

Here, we check the organisms which are currently available.

One must simply choose the ncbi_taxid and pass it as target_organism parameter to the generate_homologs function below.

show_homologene()

Convert to Orthologues and Example LIANA run

# Here, we will convert LIANA's Consensus resource to murine symbols
op_resource <- select_resource("Consensus")[[1]]

# Generate orthologous resource
ortholog_resource <- generate_homologs(op_resource = op_resource,
                                       target_organism = 10090) # mouse

# Run LIANA with the orthologous resource
liana_res <- liana_wrap(testdata,
                        resource = 'custom', # resource has to be set to 'custom' to work with external resources
                        external_resource = ortholog_resource, # provide orthologous resource
                        method=c('sca', 'natmi') # run only with sca and natmi for comp. time
                        )

# aggregate
liana_res <- liana_res %>%
    liana_aggregate()

# Plot example
liana_res %>%
    filter(source=="B") %>%
    filter(target %in% c("NK", "CD8 T")) %>%
    liana_dotplot(source_groups = "B",
                  target_groups = c("NK", "CD8 T"),
                  ntop = 15)

Conversion is trickier than anticipated. In the sense that, there will be genes without matching homologs, while others will have multiple matching homologs. These issues are further complicated when working with complexes, as is the case for the interactions in LIANA. Some ligand–receptor interactions can also be species-specific.

Thus, we suggest that if possible to check literature for support of the predicted edges when using orthology conversion.

Notably, the generate_homologs function was designed to be very flexible, but by default will be relatively conservative. So, please refer to its documentation for more information.

Generating an External Dictionary

In order to convert a human resource to any other organism, we need a dictionary which maps human gene symbols to their orthologs from another species, which might not be available via OmniPath.

An alternative widely used database for this purpose is BioMart.

An example dictionary (generated /w homologene) for murine symbols is included in liana

symbols_dict <- readRDS(file.path(liana_path, "human_mouse_orthologues.RDS"))
head(symbols_dict)

One can then simply pass the symbols_dict to the symbols_dict param to generate_homologs, instead of specifying a tax id in target_organism.

Automatic conversion comes with some limitation, we thus also refer the user to external curated databases (e.g. murine CellChatDB), which can also be used with LIANA.

saezlab/liana documentation built on Aug. 10, 2024, 8:14 a.m.