Single cell analyses require a lot of memory and often fail on the laptops. Having R + Seurat installed in a conda environment + interactive session or batch jobs with 50-100G RAM helps.

1. Use conda from bcbio

which conda
/n/app/bcbio/dev/anaconda/bin/conda
conda --version
conda 4.6.14

2. Create and setup r conda environment

conda create -n r r-essentials r-base zlib pandoc
conda init bash
conda config --set auto_activate_base false
. ~/.bashrc 

3. Activate conda env

conda activate r
which R

4. Install packages from within R

4.1 Install Seurat

R
install.packages("Seurat")
library(Seurat)
q()

4.2 Install Monocle

R
install.packages(c("BiocManager", "remotes"))
BiocManager::install("monocle")
q()

4.3 Install liger

R
install.packages("devtools")
library(devtools)
install_github("MacoskoLab/liger")
library(liger)
q()

5. Install umap-learn for UMAP clustering

pip install umap-learn

6. Deactivate conda

conda deactivate

7. (Troubleshooting)

• It may ask you to install github token - too many packages loaded from github. I generated token on my laptop and placed it in ~/.Renviron
• BiocManager::install(“slingshot”) - I failed to install it due to gsl issues.
• when running a batch job, use source activate r/ source deactivate
• if conda is trying to write in bcbio cache, check and set cache priority, your home cache should be first:
  conda info,
  ~/.condarc ``` pkgs_dirs:
  • /home/[UID]/.conda/pkgs
  • /n/app/bcbio/dev/anaconda/pkgs ```