DESeq2

DESeq2 is the standard R package for differential expression analysis of count data from RNA-seq. It models counts with a negative binomial distribution and shrinks dispersion and fold-change estimates across genes, which is what makes it reliable at the small sample sizes real experiments actually have.

What it does

The method estimates size factors for library depth, fits per-gene dispersion with information shared across genes, and tests coefficients in a generalised linear model. Shrinkage is the key idea: with three replicates per group, per-gene variance estimates are terrible on their own, and borrowing strength across genes is what keeps the false discovery rate honest.

Strengths

• Free, open source and extensively validated in the literature.
• Statistically principled handling of low replicate numbers.
• Flexible design formulae for multi-factor experiments.
• Sensible defaults with independent filtering and outlier handling.
• Exceptionally good vignette documentation.

Limitations

It requires raw counts — feeding it normalised or transformed values invalidates the model, and this remains a common and serious error. It needs R and enough statistics to specify a design formula correctly. It was built for bulk RNA-seq; single-cell differential expression has different structure and dedicated methods. Fold-change shrinkage surprises users who expect raw ratios.