Single-cell/pseudobulk Differential Analysis

Purpose and Usage

1. Single-cell Level Differential Analysis (No Biological Replicates)

Supports t-test, wilcoxon, MAST

• Scenario 1: Differential analysis between specified category 1 and category 2

  SDAS DEG -i st.h5ad -o outdir --group_key leiden --de_method wilcoxon  \
  --ident1 1 --ident2 2 \
  --fdr 0.05 --log2fc 1

• Scenario 2: Differential analysis for each category versus all others

  SDAS DEG -i st.h5ad -o outdir --group_key leiden --de_method wilcoxon \
   --fdr 0.05 --log2fc 1 

• Scenario 3: Subset a column in obs before differential analysis

  SDAS DEG -i st.h5ad -o outdir --group_key leiden --de_method wilcoxon \ 
  --ident1 1 --ident2 2 \
  --fdr 0.05 --log2fc 1 \
  --subset_key cell_type --subset_values B

2. Pseudobulk Differential Analysis (With Biological Replicates)

Recommended: DESeq2, edgeR (pseudobulk analysis). You must specify --sample_key, and the number of samples per group must meet the method requirements (DESeq2 ≥ 3, edgeR ≥ 2).

• Scenario 1: Direct differential analysis between two groups of samples

  SDAS DEG -i st.h5ad -o outdir --group_key sampleID --de_method DESeq2 \ 
  --ident1 Tumor --ident2 Normal \
  --fdr 0.05 --log2fc 1 \
  --sample_key sampleID

• Scenario 2: Subset before pseudobulk differential analysis

  SDAS DEG -i st.h5ad -o outdir --group_key sampleID --de_method DESeq2 \ 
  --ident1 Tumor --ident2 Normal \
  --fdr 0.05 --log2fc 1 \
  --sample_key sampleID \
  --subset_key cell_type --subset_values B

Input Parameter Description

DEG Parameter	Required	Default Value	Description
-i / --input	Yes		Input a h5ad file which contain gene expression matrix.
-o / --output	Yes		output directory.
--de_method	Yes		Chose a DEG method.
--group_key	Yes		Identifier name in h5ad obs, must contain ident1 and ident2.
--ident1	No		Identity class to define DEG for, if NULL, each object in --group_key will be used.
--ident2	No		A second identity class for comparison, if NULL, use the union of the rest in --group_key.
--sample_key	No		Sample key in obs (optional), must set when de_method is DESeq2 or edgeR
--subset_key	No		Key for subsetting (optional), each value will be subset for DEG if not set --subset_values
--subset_values	No		Values in --subset_key used for subsetting (optional), eg. cell1,cell2
--layer	No		Set gene raw expression layer, if NULL, adata.raw.X or adata.X will be used
--gene_symbol_key	No	real_gene_name	set gene name. default: real_gene_name for saw h5ad
--fdr	No	0.05	set adjusted p-value (FDR) cutoff to chose significant deg genes. default: 0.05
--log2fc	No	1	set absolute logfoldchanges value cutoff to chose significant deg genes. default: 1
--genelist	No	5	draw genes in volcano_plot, split genes by ',', default 5 significant genes in up and down, set 0 to not draw gene in volcano_plot
--add_label	No		Input a csv format file to add a label to obs columns
--min_gene	No	0	min genes per spot for filter, default: 1
--max_gene	No		max genes per spot for filter, default not filter
--min_cell	No	0	a gene in min cells for filter, default: 1
--volcano_xlim	No		set x limit in volcano plot, eg: -5 5.

Output Results Display

Result File	Description
de_{method}.{group_key}.{ident1}-vs-{ident2}.raw.csv	Raw output from the software
de_{method}.{group_key}.{ident1}-vs-{ident2}.all.csv	Extracted results with geneName, log2FC, Pvalue, FDR, etc.
de_{method}.{group_key}.{ident1}-vs-{ident2}.sig_filtered.csv	Significant DEGs filtered by log2FC and Pvalue
de_{method}.{group_key}.{ident1}-vs-{ident2}.png/pdf	Volcano plot in png or pdf format

• Raw file format example: de_{method}.{group_key}.{ident1}-vs-{ident2}.raw.csv This file is the original output from the differential analysis software, which may contain information such as gene name, fold change, Pvalue, adjusted Pvalue (FDR), and other details.

names	scores	logfoldchanges	pvals	pvals_adj
MTATP6P1	16.74336	1.3794351	1.3877341418899603e-42	2.2785333033340524e-39
AGR2	13.671169	1.7758344	1.419568544127444e-32	1.1147316293689464e-29
CLDN4	13.663365	1.9820584	1.9626883546881656e-34	1.6880054463820458e-31
...	...	...	...	...

• all/sig_filtered file format example: de_{method}.{group_key}.{ident1}-vs-{ident2}.all.csv This file extracts gene name, fold change, Pvalue, and adjusted Pvalue (FDR) from the original results and renames them uniformly. de_{method}.{group_key}.{ident1}-vs-{ident2}.sig_filtered.csv is the list of significant DEGs filtered by log2FC and FDR thresholds.

geneName	log2FC	pvalue	FDR
MTATP6P1	1.3794351	1.3877341418899603e-42	2.2785333033340524e-39
AGR2	1.7758344	1.419568544127444e-32	1.1147316293689464e-29
CLDN4	1.9820584	1.9626883546881656e-34	1.6880054463820458e-31
...	...	...	...

• Volcano plot result example: de_{method}.{group_key}.{ident1}-vs-{ident2}.png/pdf In the plot, red dots represent significant DEGs that meet both log2FC and FDR thresholds, blue dots meet the FDR but not log2FC threshold, and green dots meet the log2FC but not FDR threshold. By default, the top 5 up- and down-regulated genes are labeled. You can specify genes to label in the plot using the genelist parameter (e.g., --genelist geneA,geneB,geneC).

Result Interpretation

1. Gene name uniqueness

   • Before differential analysis, gene names are automatically made unique using make_unique. All outputs and plots use the unique gene names.

2. Cell and gene filtering

   • Supports filtering cells and genes using parameters such as --min_gene, --max_gene, and --min_cell. If the h5ad file has already been filtered, these can be omitted.

Parameter Tuning Suggestions

1. When the number of bins/cells exceeds 200k, MAST cannot run successfully. In this case, stricter filtering parameters (min_gene and min_cell) can be set to reduce the number of bins/cells before analysis.