Finding. Cell Painting, the high-content morphological assay behind JUMP-CP, does not sort chemicals into mammary carcinogen versus not. It sorts them by how they hit the cell. Achebouche and Taboureau took 28 mammary carcinogens and 23 non-genotoxic non-mammary carcinogens. Genotoxic mammary carcinogens shove the endoplasmic reticulum, mitochondria, nucleus, and RNA compartments. Hormone-acting compounds sit on the control cloud, because the cell line lacks working steroid receptors. Their best guilt-by-association scorer reaches AUC-ROC 0.630 and AUC-PR 0.696. Aimed at endocrine disruptors anyway, it nominates clofentezine, 3-methylpyrazole, resorcinol, 2-tert-butyl-4-methoxyphenol, and thiabendazole for a real test.
Why mitochondria are in the sentence
The organelle is a perturbed compartment on a morphological map of genotoxicity. That is useful and small. It is not a mechanism of breast cancer. The valuable honesty is the miss: endocrine disruptors are invisible here, so a mitochondria-heavy feature set will over-call DNA-damaging junk and under-call hormones.
How to read the score
Low sixties. Methods, a mitochondrial imaging channel, a weak AUC, a clear limit. Confidence is medium. Good for toxicology agents; skip for OXPHOS disease.
Caveats
Do not regulate a chemical with AUC 0.63. Do not call the five names carcinogens. Build a hormone-sensitive Cell Painting line before you claim endocrine coverage.
What to do with it
If you mine JUMP-CP for mitochondrial toxins, expect genotoxic morphology, not ER-alpha ligands. If you prioritize mammary risk, treat this as a transparent ranker with a known blind spot. Read their GBA configurations before you reimplement.