We studied endpoint mammary tumors from a genetically engineered mouse line (EIC) in which doxycycline induces HER2 together with Cre in the mammary epithelium, crossed with a conditional Ezh2 strain. In Ezh2 fl/fl mice, Ezh2 is deleted in the Cre-expressing tumor epithelium; Ezh2 wild-type (wt/wt) mice keep Ezh2 and are the controls. All mice received doxycycline from 8-12 weeks of age until their tumors reached the humane endpoint (the constant `treatment` column), and every tumor was harvested at that endpoint, so there is no time variable.

The genotype is in the `group` column: "Mouse tumor, EZH2knockout" marks tumors of Ezh2 fl/fl mice and "Mouse tumor, EZH2wildtype" marks tumors of Ezh2 wild-type mice, five tumors each. Each sample is one endpoint tumor, and the ten tumors come from ten different mice, so the samples are independent and no two come from the same animal. The replicate numbers 1 to 5 in `original_sample_id` and `geo_title` only count the tumors within each genotype; a knockout and a wild-type tumor with the same number are unrelated. Each tumor was sequenced once, so there are no technical replicates.

We have no record of processing or sequencing batches, and no other variable we consider relevant; the remaining columns are GEO annotations that are the same for all samples or only name the libraries. The data are bulk RNA-seq of RNA from snap-frozen tumor tissue, given as raw gene-level read counts (one row per gene symbol).

Our question is which genes are differentially expressed in Ezh2 knockout tumors ("Mouse tumor, EZH2knockout") compared with Ezh2 wild-type tumors ("Mouse tumor, EZH2wildtype").