Creating the R environment

Make sure you are in the folder with the code:

• analyze_correlation_results.Rmd
• correlation_workflow.Rmd

R environment

This was tested on R=4.4.1

It is optional to use Posit Package Manager. We use ubuntu 22 so we set up to:

# Configure BioCManager to use Posit Package Manager:
options(BioC_mirror = "https://packagemanager.posit.co/bioconductor/latest")
options(BIOCONDUCTOR_CONFIG_FILE = "https://packagemanager.posit.co/bioconductor/latest/config.yaml")
# Configure a CRAN snapshot compatible with Bioconductor 3.20:
options(repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/jammy/latest"))

install.packages("BiocManager")
BiocManager::install("devtools")
BiocManager::install(c("usethis", "rmarkdown", "knitr", "reticulate",
                       "reshape2", "RColorBrewer",
                       "ggplot2", "tidyverse", "glue", "gridExtra"))
BiocManager::install("pheatmap")
BiocManager::install("Seurat")
BiocManager::install("mojaveazure/seurat-disk")
BiocManager::install("SAVER")
devtools::install_github("ChangSuBiostats/CS-CORE")

usethis::use_course('https://github.com/KrishnaswamyLab/MAGIC/archive/master.zip',
                    destdir=".")
devtools::install_local("MAGIC-master/Rmagic/")

python environment

In order to run the correlation workflow, reticulate needs to be used as MAGIC is a python based tool. In order to recreate a conda environments that is compatible with running the workflow, you can create a new conda environment with requirements.txt file included in this repository. The following code can be used to create said environment:

library(reticulate)
virtualenv_create("magic2",  packages="numpy==1.26",python_version="3.9")
virtualenv_install("magic2", packages="magic-impute")
#.rs.restartR() # restart R manually if outside RStudio

User supplied inputs

After installing the needed R packages as well, you can being working with the correlation_workflow.Rmd to generate a HTML report. The following pieces of information need to be provided in the section titled "User Inputs":

• Path to the seurat object: path_seurat
• Directory where intermediate and final results will be stored: path_outs
• Name of the metadata column where celltypes are stored: col_celltype
• Celltype of interested (that will be subset to): ct
• Boolean (TRUE/FALSE) filter value on whether or not to filter genes based on expression and frequency (if TRUE, will remove genes based upon various thresholds): filter
• Minimum average expression for a gene (Default 0.2): min_exp
• Minimum number of cells a gene must be expressed in (Default 40): min_cells
• Minimum percent of cells a gene must be expressed in (Default 0.2): min_perc
• List of genes the calculate correlations between: corr_genes_all

Outputs

When the correlation_workflow.Rmd is knit, a HTML file of the same name will be generated. This report will have the following information:

• Basic information about the seurat object provided
• Summary of gene expression for all genes of interest
• Heatmaps showing the correlation estimates for every gene pair
• Compare the correlation scores and the significance between SAVER, CS-CORE, and MAGIC

Additionally 3 other files will be generated in the folder where path_outs was specified:

• imputed_{ct}.RDS: Seurat object with assays including the counts for:
  • log-normalization (RNA)
  • SCT
  • MAGIC
  • SAVER
• corr_{ct}.csv: Table of all correlation scores for each method
• p_val_{ct}.csv: Table of all p-values from the correlation calculation for each method