A sleuth scanned 270,000 antibody catalogue images. More than 18,000 appear edited.
A metascientist at Northwestern University scanned more than 270,000 antibody catalogue images and flagged over 18,000 as apparently altered, across 15 commercial suppliers. The most telling detail is not the total but the backgrounds: some 13,800 of the flagged images share just seven of them, a pattern one image-integrity specialist says implies the antibodies were never tested in a laboratory at all. The finding moves research-integrity scrutiny one step upstream of the published paper, into the catalogue a researcher reads before the experiment exists.
On 25 August, Reese Richardson, a metascientist at Northwestern University in Evanston, Illinois, reported that he had found more than 18,000 questionable images attached to antibodies sold by 15 commercial suppliers, and released a public database of them. Nature reported the finding the same day.
The method was blunt and reproducible. Richardson wrote a program that raised and lowered the black and white levels across more than 270,000 catalogue images to increase contrast and expose editing, then checked each flagged image by hand. Four vendors account for most of the total: G-Biosciences in St. Louis, Missouri (6,388 images), Thermo Fisher Scientific in Waltham, Massachusetts (5,580), LSBio in Newark, California (2,695) and Origene in Rockville, Maryland (2,236). Most other vendors in the database are associated with a few hundred each.
Richardson led the earlier, much smaller inquiry in May that found more than 100 suspicious images in Thermo Fisher's catalogue.
The pattern that matters more than the count
The headline number is 18,000. The more informative figure is 13,800: that is roughly how many of the flagged images share one of only seven backgrounds.
Image backgrounds in a western blot are normally random, closer to television static than to a template. A shared background is not a contrast adjustment or a crop. In many cases the same background appeared across images from different vendors.
"That particular type of image alteration or manipulation suggests that these antibodies were never tested at all in the lab."
That is Elisabeth Bik, a specialist in image integrity, quoted by Nature. It is a claim about the existence of an experiment, not about its presentation, and it is the reason this story is not simply about tidy figures.
What the companies say
- Thermo Fisher Scientific: an ongoing review has so far found "no evidence of impact to antibody performance or the underlying scientific data", and the company remains "confident in the quality of the products within our antibody catalog".
- LSBio: the entries "amount to less than 0.04% of the total offering on our platform", but the firm takes the concerns seriously and will investigate with its partners.
- G-Biosciences and Origene: did not respond to Nature's request for comment.
Image alteration does not by itself mean a product is defective. That distinction is worth holding onto, and Simona Colantonio, who directs the Antibody Characterization Laboratory at the Frederick National Laboratory for Cancer Research, made the same point from the other direction: "The sad part is some of the antibodies are actually probably very good", and the alterations cast doubt on them anyway.
A context figure worth reading carefully
Nature's report notes a 2023 study of 614 commercial antibodies in which more than 50% "did not work as advertised". The underlying paper is worth opening rather than quoting second hand.
Ayoubi and colleagues tested 614 commercial antibodies raised against 65 neuroscience-related proteins. Of the 578 that manufacturers recommended for western blot, 44% performed as intended, 35% bound their target but were not selective, and 21% failed outright (eLife 12, RP91645).
So the "more than 50%" is the second and third categories added together, for one application. It is accurate. It is also a threshold placed on a distribution the authors published in full, and the distribution is the more useful object: "did not work" covers both an antibody that binds nothing and an antibody that binds its target plus several other things. Those two failures mislead a reader in different ways.
Analysis
Research-integrity work has concentrated on the published record: the figure in the paper, the retraction notice, the sleuth reading a supplementary file. This is a flag raised one step upstream, in the sales catalogue a researcher consults before the experiment exists.
That changes who is accountable and what a correction would even look like. A journal can retract. A vendor catalogue has no equivalent instrument, no notice of record, and no obligation to tell the thousands of laboratories that already bought on the strength of a validation image. Sholto David, a molecular biologist who works on research-integrity issues, told Nature that catalogue images are "really one of the few things you have to go on" when choosing an antibody. If that signal is unreliable, the appraisal a careful researcher performs at the bench begins from a corrupted input, and no amount of downstream rigour recovers it.
Harvinder Virk, co-founder of the open-science effort Only Good Antibodies, framed it as "a system-wide problem that requires a system-wide solution". The reproducibility literature already contains the shape of one: independent characterisation, published openly, rather than validation supplied by the party selling the reagent.
What to watch
Richardson's 18,000-plus flags have not all been independently double-checked, a limitation Nature states plainly and which the researchers it consulted did not dispute; they judged the method sound rather than the individual verdicts confirmed. Independent replication of the scan, and whether any vendor issues corrections to specific catalogue entries rather than a general statement of confidence, are the next things worth following.