Benchmark Pipeline for CLIC scores

Overview

The CLIC package provides the tool for feature selection for integrating unpaired scRNA-seq and scATAC-seq data in a way that leverages multiome data from ENCODE.

CLIC package can be found here

This repository contains all of the scripts used to evaluate CLIC against baseline variable feature selection when used within the workflow of SOTA methods.

For more details about the algorithm as well as the benchmarking process, please refer to our preprint: PREPRINT LINK PENDING

Description

The scripts directory contains all code used to generate benchmark results:

• scripts/preprocessing/
  Scripts for preprocessing raw data downloaded from original publications.

• scripts/methods/
  Scripts for running each integration algorithm (Seurat, GLUE, Liger, bindSC) with either CLIC-based or baseline variable feature selection.

• scripts/benchmark/
  Contains SLURM job submission scripts and parameter configurations for running benchmark experiments on HPC:

  • submit_{paired, unpaired, spatial}.sh and submit_gpu_{paired, unpaired, spatial}.sh: submission scripts for integration methods.
  • parameters/: code for generating parameter combinations used to assess robustness across settings.
  • run_benchmark_{'', 'unpaired', 'spatial'}: scripts for computing key metrics to evaluate integration quality.