Differential Expression Analysis

The fundamental metric of effect size in differential expression. It represents the logarithmic ratio of normalized expression between two conditions, centered at zero. Positive values indicate up-regulation, negative values indicate down-regulation. The log2 transformation symmetrizes fold changes: a 2-fold increase is +1, while a 2-fold decrease is -1. This symmetry is critical for downstream visualization in volcano plots and for setting biologically meaningful significance thresholds.

The expected proportion of Type I errors among all rejected null hypotheses. In a typical RNA-seq experiment testing 20,000 genes, an uncorrected p-value cutoff of 0.05 would yield ~1,000 false positives. FDR control, typically set at 5%, limits this to a manageable number. The Benjamini-Hochberg procedure is the most common implementation, ranking raw p-values and comparing each to an adjusted significance threshold. FDR is the standard for genomic studies where controlling false positives is paramount.

The statistical backbone of modern RNA-seq differential expression tools. Unlike the Poisson distribution, which assumes the variance equals the mean, the negative binomial explicitly models overdispersion — the biological reality that gene expression variance exceeds the mean due to biological and technical variability. Both DESeq2 and edgeR use this distribution to accurately model count data, preventing the inflated false positive rates that plague Poisson-based methods.

A statistical stabilization technique that borrows information across all genes to improve estimates for individual genes. Genes with low read counts or high variability have unreliable dispersion and fold change estimates. Empirical Bayes methods shrink these noisy estimates toward a common prior distribution derived from the entire dataset. This prevents genes with small counts from dominating ranked gene lists with artificially large fold changes, a critical safeguard in DESeq2 and limma.

The canonical visualization for differential expression results. It plots -log10(p-value) on the y-axis against log2 fold change on the x-axis for every gene. This dual-threshold view simultaneously highlights both statistical significance and biological effect size. Genes appearing in the upper-left and upper-right corners are the most promising candidates — highly significant with large magnitude changes. Modern interactive volcano plots allow brushing to identify gene symbols and link to functional annotation databases.

Systematic non-biological variation introduced by processing samples in different groups, on different days, or by different technicians. Uncorrected batch effects can completely confound biological signals, leading to spurious discoveries. ComBat-seq uses a negative binomial regression framework to adjust for known technical covariates while preserving the integer nature of count data. Proper experimental design with balanced batch allocation remains the most effective mitigation strategy.