Comprehensive benchmark of dimension reduction methods for single-cell data analysis

Software requirements

• Nextflow
• Docker

The list of benchmarking datasets

• scATAC-seq:

How to prepare new benchmark datasets

Benchmark data is stored in anndata format. The anndata object should contain:

1. adata.X: cell by feature count matrix. For scATAC-seq, the format of feature names should be: chr:start-end.
2. adata.obs["cell_type"]: cluster/cell-type annotation of each barcode.
3. data.obs['batch']: batch annotation of each barcode (optional).
4. data.obs['compare']:differentially accessible region between two groups ref vs case
5. Groundtruth have at least two columns: ['cell_type'] and DARs named by ['index']

How to run benchmark

The pipeline uses docker containers to run the benchmark. Therefore, you need to install either docker on your machine. The pipeline needs to download large data files, so make sure you have enough disk space (>20G) and a fast internet connection.

Use ./bench.sh -profile singularity or ./bench.sh -profile docker to run benchmarks. The benchmark results will be stored in ./results folder.

See nextflow.config for additional configuration options.