statLens Method Card
DESeq2: negative binomial GLM with shrunken dispersions and fold changes
v1.52.0STATLENS VERIFIEDTests for differential expression in sequencing count data with a negative binomial GLM per gene. Dispersions and fold changes are shrunk with empirical Bayes, which keeps estimates stable for low-count genes and small sample sizes.
Category
Differential abundance
Analysis task
Differential analysis
Study designs
Cross-sectional, Case-control, Cohort, Paired, Clinical trial, Multi-centre
Source
AI-drafted from the paper and package documentation; pending expert review
Scientific questions this method addresses
Which genes are differentially expressed between conditions in bulk RNA-seq count data?
How large is each gene's expression change, with noisy low-count estimates shrunk so genes can be ranked?
Which genes respond to treatment after adjusting for batch or for paired donors?
Statistical specification
K_{ij} \sim \mathrm{NB}(\mu_{ij}, \alpha_i), \quad \mu_{ij} = s_j q_{ij}, \quad \log_2 q_{ij} = x_j^\top \beta_i
Model family
Generalized linear model per feature with empirical Bayes shrinkage
Distribution
Negative binomial
Estimand
log2 fold change per gene; H₀: β_ir = 0 (Wald test), or a reduced model (likelihood ratio test)
Multiple testing
Benjamini–Hochberg FDR
Assumptions
Input is raw integer counts; sequencing depth differences are captured by size factors; Most genes are not differentially expressed (median-of-ratios normalization); Genes with similar mean counts have similar dispersion (dispersion–mean trend); Samples are independent given the design; only fixed effects are modelled
Sample size
At least 2 replicates per group; 3 or more recommended
Best used when
Bulk RNA-seq with few replicates (3 or more per group)
Ranking genes by shrunken fold change (lfcShrink)
Designs with batch or paired donor as fixed effects
Other count assays with the same structure, such as ChIP-seq, ATAC-seq or CRISPR screen counts
Do not use when
Data are already normalized, transformed or non-integer
Repeated measures need random effects (DESeq2 fits fixed effects only)
Most features are expected to change, which breaks median-of-ratios normalization
Sparse single-cell counts are analysed with default settings (see the vignette's single-cell recommendations)
Relationships
Typed edges to other Method Cards. These are what let statLens assemble multi-step workflows rather than single methods.
supersedesDESeqAdds fold-change shrinkage and improved dispersion estimation
alternative_toedgeRSame negative binomial family; differs in normalization (median-of-ratios vs TMM) and in shrinkage of fold changes
alternative_tolimmalimma-voom transforms counts and fits weighted linear models; faster with large sample sizes
Outputs
Mapped to the statLens standard statistical result, plus method-specific fields.
Standard fields
feature_idestimatestandard_errortest_statisticp_valueadjusted_p_value
Method-specific
baseMeandispersion