Evaluating batch correction methods for image-based cell profiling

This repository contains the source code to reproduce the results in the paper: "Evaluating batch correction methods for image-based cell profiling".

These scripts generate several embedding visualizations and .csv files for quantitative evaluation of batch correction methods.

Installation

We suggest Mamba for environment management. The following commands create the environment from scratch and install the required packages.

mamba env create --file environment.yaml
mamba activate batchcp

kBET installation

Run the following command to install R package kBET:

R -e "remotes::install_github('theislab/kBET')"

scib troubleshooting

The single-cell integration benchmark scib package may fail because it includes c/c++ code that should be compiled specifically for your environment. An alternative is to install it from source:

DEST=$HOME/projects/scib
git clone https://github.com/theislab/scib.git $DEST
cd $DEST
git checkout v1.1.4
pip install -e .

More info in this issue

mnnpy installation

Similarly, mnnpy may require manual installation. More info at https://github.com/chriscainx/mnnpy#install

Get input data

Download profiles and metadata:

source download_data.sh

Run scenarios

Every scenario reported in the paper can be reproduced running snakemake with the associated config file. For example, to reproduce Scenario 1 using 3 cores:

snakemake -c3 --configfile inputs/conf/scenario_1.json

You can get the scores, corrected profiles and plots in the ./outputs folder.