CHOIR improves significance-based detection of cell types and states from single-cell data

Brief descriptions of analysis scripts

Pre-processing pipeline (See preprocessing_scripts folder)

This folder contains code to run a standardized pre-processing pipeline that was used to perform read alignment, quality control, and normalization steps on the datasets used in the benchmarking assessments.

Clustering methods (See clustering_scripts folder)

This folder contains code to run 16 different clustering methods, including CHOIR, using a series of parameter combinations, and compiles the resulting cluster labels, computational time, and memory usage. Also contains scripts to run clustering using ArchR, Signac, scTriangulate, and the CHOIR functions specific to atlas-scale data.

Benchmarking analysis for simulated datasets (See additional_analysis_scripts/simulated_datasets folder)

generate_simulated_data.R - Code to generate the simulated datasets for clustering method benchmarking using Splatter.

generate_simulated_data_modulated.R - Code to generate the additional simulated datasets with various features modulated using Splatter and scDesign3.

simulated_data_metrics.R - Code to assess benchmarking metrics for the clustering of each simulated dataset.

simulated_data_plots.R - Code to generate plots for the analysis of the simulated datasets. See Supplementary Figs. 3–24 and 28–30.

simulated_data_plots_modulated.R - Code to generate plots for the analysis of the simulated datasets with various features modulated. See Supplementary Figs. 25–27.

simulated_data_compiled_metrics.R - Code to compile benchmarking metrics across the 100 simulated datasets. See Fig. 2.

run_subclustering.R - Code to run the subclustering analysis on Simulated Datasets 46–50. See Extended Data Fig. 5.

subclustering_analysis_metrics.R - Code to assess the benchmarking metrics for the subclustering analysis of Simulated Datasets 46–50. See Extended Data Fig. 5.

Benchmarking analysis for real single-cell sequencing datasets (See additional_analysis_scripts/real_datasets folder)

Wang_2022_10x_Multiome_human_retina.R - Code to analyze and generate plots for the Wang et al. 2022 10x Multiome human retina dataset. See Extended Data Fig. 2 and Supplementary Fig. 1.

Siletti_2023_human_brain_atlas.R - Code to analyze and generate plots for the Siletti et al. 2023 human brain atlas dataset. See Extended Data Fig. 3.

Kinker_2020_cancer_cell_lines.R - Code for the benchmarking analysis of the Kinker et al. 2020 scRNA-seq cancer cell line dataset. See Figs. 3–4 and Supplementary Fig. 31.

Hao_2021_CITEseq_human_PBMCs.R - Code for the benchmarking analysis of the Hao et al. 2021 CITE-seq human PBMC dataset. See Fig. 5 and Supplementary Fig. 32.

Srivatsan_2021_sciSpace_mouse_embryo.R - Code for the benchmarking analysis of the Srivatsan et al. 2021 sci-Space mouse embryo dataset. See Fig. 6, Extended Data Figs. 6–10, and Supplementary Figs. 33–34.

MAGIC_imputation.R - Code to run MAGIC imputation for selected genes.

Data availability

Simulated datasets generated using R package Splatter

Simulated Datasets 1–100 - https://files.corces.gladstone.org/Publications/2024_Petersen_CHOIR

Publicly available datasets used in the benchmarking assessments

Wang et al. 2022 10x Multiome human retina data - GEO accession number: GSE196235

Siletti et al. 2023 human brain atlas data - No GEO accession number. See publication

Kinker et al. 2020 scRNA-seq cancer cell line data - GEO accession number: GSE157220

Yang et al. 2021 scRNA-seq A375 cancer cell line data - GEO accession number: GSE164614

Dave et al. 2023 scRNA-seq T47D cancer cell line data - GEO accession number: GSE182694

Hao et al. 2021 CITE-seq human PBMC data - GEO accession number: GSE164378

Srivatsan et al. 2021 sci-Space whole mouse embryo data - GEO accession number: GSE166692

Note 1. File paths in some of the provided scripts are hard-coded.