Why is normalization required for RNA-seq count data?

Why

A sample with twice as many mapped reads can show about twice the count for the same transcript despite identical underlying biology. Methods such as DESeq2 size factors or edgeR TMM estimate scaling using assumptions about the distribution of gene changes. This differs from comparing TPM values, which correct for transcript length and library composition but do not automatically supply suitable inputs for count-based tests.

Imagine two treated cultures with globally different abundance profiles. A handful of highly abundant transcripts can distort naïve total-count scaling, so inspect composition before choosing a method. For differential gene analysis, the model usually starts from raw or suitably estimated counts and includes offsets or normalization factors rather than treating transformed expression as independent observations.

Normalization removes certain technical scale differences. It cannot fix poor RNA integrity, variable cell composition, wrong annotations, or treatment that is perfectly confounded with library batch. Explain which biological quantity the method estimates and inspect QC plots before interpreting significance.