Pre-fitted CellMentor model with balanced human PBMC reference and query (companion data for Petrenko & Aran, Methods in Molecular Biology, 2026)

CellMentor performs supervised, cell-type-aware non-negative matrix factorization. It learns a gene-by-factor matrix W from an annotated reference, then projects unannotated query cells onto that same factor space. Learning W is the expensive step; projecting onto it is cheap. This deposit provides the reference, the query and W already fitted, so the chapter's worked example can be reproduced without repeating a multi-hour decomposition.