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

knitr::opts_chunk$set(echo = TRUE)

LIANA ++

To facilitate further method development and the benchmarks of current and future methods, in this vignette we provide further instructions relevant for method developers.

Load prerequisites

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

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

liana_pipe and Contributions

To facilitate method development, we provide a guide how to make use of the architecture already implemented in LIANA.

First, we start by converting the object to SCE (if Seurat) and basic checks via liana_rep. Then we generate a LR_res table with many relevant columns, such as stats for ligands and receptors per cell type. Note that liana_pipe also takes complexes into account by dissociating them into subunits (see ?liana:::decomplexify). These are later re-assembled back into complexes using the recomplexify function.

sce <- liana_prep(testdata)

# Obtain LR summary res with CellPhoneDB and dissociate complexes
lr_res <- liana_pipe(sce,
                     op_resource = select_resource("CellPhoneDB")[[1]] %>%
                       liana:::decomplexify(),
                     base = exp(1) # here we assume log-transformation
                     # of the library-normalized counts
                     ) %>%
    glimpse()

``

The LR result summary is then passed to the get_* functions then provide ways to summarize this information into a score (LRscore in this case) that can be used to the prioritize interactions.

We also need to pass the function that we wish to use to account for heteromeric complexes

get_sca(lr_res, complex_policy="mean",
        expr_prop=liana_defaults()[["expr_prop"]],
        return_all=FALSE) %>% glimpse

get_* functions refer to liana_call and liana_score functions behind the scenes, so please refer to those for more info.

Install LIANA++

Install a LIANA++ environment with all original software as a way to minimize bias for benchmark studies.

LIANA++ uses CCC methods from both R and Python, as such dependencies for both need to be installed.

To install the LIANA++ framework run the following code in R:

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

if (!requireNamespace("devtools", quietly = TRUE))
    install.packages("devtools")

library(devtools)

Sys.setenv(R_REMOTES_NO_ERRORS_FROM_WARNINGS = TRUE) # ignore warning from iTALK
BiocManager::install("ComplexHeatmap") # required for Connectome
devtools::install_github('saezlab/OmnipathR@ff3ad88e3915747e1b557bf44ac5396f9525dd7e') # install 4.0 version of OmnipathR

# install tools
devtools::install_github("sqjin/CellChat")
devtools::install_github('msraredon/Connectome', ref = 'master')
devtools::install_github("Coolgenome/iTALK", build_vignettes = FALSE)
# A modified version of SingleCellSignalR (SCA) that enables external resources
devtools::install_github(repo = "saezlab/SingleCellSignalR_v1",
                         subdir = "SingleCellSignalR")

# Finally, install LIANA
devtools::install_github('saezlab/liana')

Squidpy and NATMI are written in Python, as such a python environment with the prerequisites of these methods needs to be set up.

If you also wish to run the CellPhoneDB re-implementation from Squidpy, please set up a conda environment by running the following lines in the terminal: ```{bash, eval = FALSE} conda create -n liana_env conda activate liana_env conda install -c anaconda python=3.8.5 pip install squidpy

Please use the [.yml](https://github.com/saezlab/liana/blob/master/liana_env.yml) file to set up a conda environment:

conda env create -f liana_env.yml

Note! NATMI and Squidpy are set by default to look for this conda environment ("liana_env").

#### To use NATMI with LIANA, clone the modified NATMI repo into liana path by running the following in the terminal:
```{sh, eval=FALSE}
cd *insert fullpath*/liana
git clone https://github.com/saezlab/NATMI

Note that, NATMI was forked from asrhou/NATMI and changed to be agnostic of the working directory when loading the resources.

If needed, use the following to locate the liana package

system.file(package = "liana")
# "/home/user/R/x86_64.../4.0/liana"

It might be easier to just also clone the liana repo.

call_* functions

Use the call_ functions to run the pipelines of the original methods, rather than their re-implementations. These functions work with Seurat objects and they are instead converted to SCE. These are largely getting deprecated and will be moved to a seperate repository in the future.

# RUN cellchat alone with OmniPath's Consensus
sca_res <- call_sca(
  testdata,
  op_resource = select_resource('Consensus')[[1]]
  )

# Show CellChat Results
sca_res

Similarly one could call all call_* functions via liana_wrap

liana_calls <- liana_wrap(testdata,
                          method = c('call_connectome',
                                     'call_sca',
                                     'call_italk',
                                     'call_natmi'
                                     ),
                          resource = c("Default")) # Run with Default Resources
liana_calls %>% liana_aggregate() # Connectome finds no significant hits here
# Note that for italk we use mean(ligand.logFC, receptor.logFC) as a way to compare
# to the other methods

Session information

options(width = 120)
sessioninfo::session_info()

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