18,943 images that were supposed to be proof
Scientists buy antibodies on the strength of a picture. On August 25, a database said 18,943 of those pictures don't add up.
It started in May: Sholto David found one faked blot at Thermo Fisher. Richardson scanned 270,000 images, boosted contrast, checked each by hand. The pattern: thousands of blots sharing one background, bands flipped and pasted, blemishes painted over. One background recurs across six vendors. Elisabeth Bik, the field's best-known sleuth: some of these antibodies were likely never tested at all.
Thermo Fisher, the world's largest lab supplier: 5,580 flagged images. Its answer: stands by the data; images may have been 'adjusted for presentation.' G-Biosciences, linked to 6,388, said nothing. LSBio called its share 0.04% of its catalog.
Fair caveat: a doctored image does not prove the antibody fails. It proves you can't trust the proof. Scientists choose reagents by these pictures. A 2023 study found over half of 614 commercial antibodies didn't work as advertised; the bill is estimated past a billion a year.
Richardson: not a complete audit; a vendor absent from it is not a clean vendor. The checkers are volunteers. The checked are industry giants.
Checked against the public record: the database itself, Nature's news desk, C&EN, Chemistry World.