Method and sources

How this list was built, and what it does not say

This site republishes a public, citable dataset of antibody catalogue images that image-forensics contributors have flagged as manipulated, duplicated or otherwise unreliable, and joins it to independent knockout-controlled test results where those exist.

Where the data comes from

On 17 May 2026 Sholto David identified a single fabricated image presented as validation data in Thermo Fisher’s antibody catalogue. By 25 August 2026 the community repository maintained by Reese Richardson held 18,944 images across 17,495 products from 16 vendors. Nature reported the expansion on the same day.

Every row on this site comes from that repository, version 260825, which is published under CC BY 4.0. Nothing here is our own image analysis. The annotations shown on each product page are the contributors’ own.

What “flagged” means

The dominant findings are not subtle. More than 7,500 images share one identical field of background noise (“pattern A”); thousands more share patterns B through H. Independently acquired images cannot share a noise field. A further 4,700 images show artefacts of painting — regions brushed over to remove unwanted features. Others are the same blot sold under different catalogue numbers, by the same vendor or by several vendors at once, which is consistent with private labelling: one manufacturer, several brands, one set of validation data passed along.

What it does not mean

• A flag is not proof that an antibody fails. It is proof that the published evidence is not evidence. The right response is to treat the product as untested.
• Fewer flags for a vendor does not mean a cleaner catalogue. No catalogue was audited completely, and the review effort was not spread evenly.
• A vendor’s absence from the list is not a clean bill of health.
• Images too low in resolution for a confident judgement were deliberately excluded, so the true number is higher than the number shown here.
• Responsibility for a manipulated image is not established by its presence in a catalogue. Several vendors resell products, and validation data, produced by others.

The independent check

Only Good Antibodies publishes knockout-controlled characterisation data from YCharOS: the antibody is run against a genetic null control, and the raw figures are released. 52 of the flagged products appear in that dataset — 0.3 % of the list. Of those, 20 failed in every application tested. The rest work at least partly. A manipulated catalogue image, on its own, does not predict the outcome; it only removes the ability to know.

If you are using one of these antibodies

Do not discard it on the strength of this list. Control it: a knockout or knockdown lysate, a peptide competition, or a second antibody raised against a different epitope. Then publish what you find — the reason this problem stayed invisible for so long is that negative validation results rarely get written down.

Cite the primary sources, not this page

1. Richardson, R. et al. Problematic images in vendor antibody verification data, version 260825. Zenodo. doi.org/10.5281/zenodo.20402475 (CC BY 4.0)
2. More than 18,000 questionable images found in antibody catalogues of 15 companies. Nature news, 26 August 2026. nature.com/articles/d41586-026-02635-w
3. Richardson, R. At least 15 companies are selling antibodies using faked validation data. 25 August 2026. reeserichardson.blog
4. Ayoubi, R. et al. Scaling of an antibody validation procedure enables quantification of antibody performance in major research applications. eLife (2023) — the study in which more than half of 614 commercial antibodies did not perform as advertised. doi.org/10.7554/eLife.91645
5. Only Good Antibodies. Knockout-validated characterisation data. onlygoodantibodies.co.uk

Reporting and corrections

New suspicious images belong in the primary repository, not here — Richardson maintains a submission form for that. If a vendor has removed or replaced an image, or a row on this site is wrong, write to the Schlein Lab and it will be corrected or removed.

Reuse

The underlying dataset is CC BY 4.0; so is this rendering of it. The full table is available as CSV and through a read-only JSON API (/api/search?q=&vendor[]=&problem[]=&ko=).