Differential Expression

Differential expression

Methods

Method	Model	Requires	Notes
Moderated t (eBayes)	Linear model on log₂(CPM + 1) per gene, empirical-Bayes variance moderation (limma lmFit + eBayes)	Any input	Default. Supports covariates.
voom	Precision weights from the mean–variance trend, then the moderated t (limma voom)	Raw counts	Recommended for counts, particularly when library sizes vary.
Welch t	Unequal-variance t-test per gene	Any input	No variance moderation; low power at small n. Provided for comparison.

All three report log₂ fold change, t, p and Benjamini–Hochberg FDR. A gene is called significant when FDR is at or below the chosen level and |log₂FC| is at or above the chosen threshold.

Which samples enter the model

A contrast between two groups is fitted on the samples of those two groups. The genome-wide moderated F test (Group Patterns tab) uses all groups. See Limitations for the consequences.

voom details

The implementation follows limma 3.66, including two details that affect results:

1. Genes with zero counts in all samples of the contrast are excluded from the trend fit.
2. The lowess span is chosen adaptively (chooseLowessSpan, the limma 3.66 default), not fixed at 0.5.

The fitted trend is shown in the voom mean–variance trend panel.

Diagnostics

• p-value histogram: a uniform distribution with a spike near zero is expected. A U-shape or a spike near one suggests a model problem.
• Q–Q plot: observed against expected −log₁₀ p.
• Stability curves: the number of hits across FDR and fold-change thresholds, with the current thresholds marked. A result that depends on one exact threshold is fragile.

Volcano and MA plots

Top genes can be labelled, ranked by p-value (conventional), by |log₂FC|, or by π score (|log₂FC| × −log₁₀ p). The ranking used is stated in figure captions.

DESeq2

benchsiDE does not reimplement DESeq2. DESeq2 script (R) exports an R script that runs DESeq2 on the original files for the current contrast.

References

Robinson MD and Oshlack A (2010) Genome Biol 11:R25. Smyth GK (2004) Stat Appl Genet Mol Biol 3:Article3. Law CW et al. (2014) Genome Biol 15:R29. Ritchie ME et al. (2015) Nucleic Acids Res 43:e47. Benjamini Y and Hochberg Y (1995) J R Stat Soc B 57:289.