More Than 18,000 Images in Antibody Catalogs from 15 Companies Show Signs of Editing
Reese Richardson compiled an open database of 18 943 images from commercial antibody listings that show signs of possible editing
On 25 August, metaresearcher Reese Richardson published a database of observations on catalog images in the open Zenodo repository. It contains records for 17 495 products from 16 companies. Each record links to the product page, the original image, an annotated version, and a brief description of the observation.
An antibody is a laboratory reagent that binds to a specific protein. Suppliers often include a western blot on the product page. In this assay, proteins in a sample are separated, and the antibody produces a signal that appears as a band. Researchers use these images as one source of evidence when choosing a reagent for a subsequent experiment.
In a Nature investigation, Richardson described how he expanded his search to other catalogs after finding more than one hundred suspicious images in the Thermo Fisher catalog in May. His software adjusted the black and white levels in more than 270 000 images to increase contrast and reveal possible traces of editing. Richardson then reviewed the images manually. Among the flagged images, one of seven identical backgrounds appeared approximately 13 800 times.
“Some of them are so egregious that the signs of manipulation are visible even without increasing the contrast,” said Simona Colantonio, head of an antibody validation laboratory, in an interview with Nature.
The database allows each flag to be checked directly. Every row lists the supplier, catalog number, product name, product page link, original image, and annotated image. Researchers can open an individual record and compare the image with the annotation and the description of the observation before selecting a reagent for an experiment.
The database includes two Thermo Fisher listings for p16INK4a, a protein used as a marker of cellular senescence, a state in which a cell stops dividing. In an analysis of confusion between p16INK4a and a similarly named cytoskeletal protein, the catalog number established which protein the antibody was intended to recognize. In a review of 54 papers on cellular aging, the methods sections listed antibodies against bacterial β-galactosidase instead of the mammalian protein. Researchers can now compare a catalog listing with the original image, its annotation, and the description of the observation before ordering the reagent.