Over 18,900 problematic images in antibody catalogs

A Zenodo repository published its August 25 snapshot with 18,943 images attached to 17,495 antibody products sold by 16 companies. Data collectors Reese Richardson and Sholto David describe the figures as suspicious, apparently manipulated, or otherwise problematic verification material found in research-reagent catalogs.¹

Those totals measure the repository, not the share of the antibody market affected. The collection accepts outside submissions, does not draw from a representative sample, and is not a complete audit of any vendor's catalog.² A company with more entries has not necessarily produced a higher rate of questionable images.

An annotated figure also falls short of proving fraud or showing that its antibody does not work. It points to a visual pattern that should be checked against source files, editing history, and fresh product validation.

The snapshot links each image back to a product

The deposit includes a spreadsheet with the vendor, catalog number, product page, and a description of the suspected issue. A 526.4 MB archive stores the images as downloaded from vendor sites; a separate 1.1 GB archive contains annotated copies.¹ The structure makes it possible to locate each figure, compare entries, and return to the commercial page.

The description and README report 18,943 images across 17,495 products. The downloadable CSV contained a one-row discrepancy when accessed on August 26: its index reached 18,944, with one additional Thermo Fisher row, while the product total stayed unchanged. The 18,943 figure is the total declared by the versioned record and its vendor table.¹

Richardson's same-day post names 15 vendors.² The updated Zenodo record lists 16 companies, making the archived version the source for the snapshot totals. That difference is one reason the date and version belong beside any count.

Catalog images influence which reagent a lab buys

Commercial antibodies are reagents for finding a particular protein in a sample. Researchers rely on catalog data to choose a product and decide where it may work. Thermo Fisher's own validation page says poor specificity or application performance can prevent good results and cause delays, while presenting gallery data as evidence supporting confidence in a reagent.⁴

A Western blotA technique that separates proteins and uses antibodies to reveal bands on a membrane. separates proteins by size, transfers them to a membrane, and exposes that membrane to an antibody. Binding appears as bands. A band at the expected position can support target recognition; additional bands may show that the antibody is also binding other proteins. A cleaner-looking blot therefore creates a stronger impression of specificity.

One rigorous control compares ordinary cells with an isogenic knockout line whose target gene has been removed. The correct band should disappear in the knockout. A reviewed eLife study applied standardized tests to 614 commercial antibodies for 65 proteins and published the underlying data. More than half failed in at least one application. Of 578 products recommended by manufacturers for Western blot, 44% met the study's success criterion, 35% found the target while also recognizing unrelated proteins, and 21% failed.⁵

That independent study establishes why application-specific validation matters. It did not test the 17,495 products in the image repository and cannot supply a market-wide failure rate.

Repeated backgrounds dominate the annotations

Thousands of spreadsheet rows say a blot's background matches one of a small set of recurring patterns. Other entries describe apparently painted regions, duplicated blocks of background noise, or abrupt boundaries consistent with splicing. The annotations include identical bands after rotation or flipping and whole images used for different antibody clones.¹ ²

Some examples appear to place the same band at different molecular-weight positions on a shared background. The repository also has a smaller number of fluorescence images described as reused or missing cells present in another copy.¹ These are the collectors' visual assessments. Full-resolution source data are still needed to confirm what happened.

The Office of Research Integrity's guidance for scientific images treats simple whole-image adjustments and crops that preserve context as generally acceptable, with the original file retained. Selective changes such as cloning or hiding regions alter the data a reader can evaluate.⁷

Thermo Fisher's current page for its SF1 antibody says the company is reviewing Western blots in the catalog, including third-party material. It warns that images not marked “raw-unedited” may have been edited, optimized, or otherwise adjusted for presentation.³ That notice belongs to Thermo Fisher's review. It does not speak for the other 15 companies or resolve each repository entry.

Source files determine what the picture can prove

Image-data specialist Mike Rossner told The Scientist that examples Richardson presented showed clear manipulation, while cautioning that a published figure alone cannot establish intent or fabrication without the source data.⁶

The product claim has another layer. A genuine band moved onto a different background misrepresents the apparent quality or specificity of the antibody, even if the reagent still binds its target. Examples in which a band seems copied between different experiments raise deeper doubts about whether the validation was performed. An individual audit and repeat experiment remain necessary before declaring the antibody itself invalid.⁶

Vendors may buy antibodies and their supporting data through private-label arrangements, so identical images across catalogs do not by themselves identify who made an edit. Original files, acquisition metadata, intermediate versions, and supplier records can establish provenance. A knockout control or another independent assay then measures performance in the intended application, which visual forensics cannot do on its own.

Vendor responses are not interchangeable

Abcam and MilliporeSigma told The Scientist they were reviewing catalog imagery. LSBio said it would investigate the examples sent by the publication and examine source data. Thermo Fisher, ProteoGenix, and Origene had not answered that outlet's requests for comment when its story was published.⁶ The notice currently visible on Thermo Fisher's page records an announced review without settling the items in the Zenodo collection.

Sources

1. Problematic images in vendor antibody verification data, version 260825 · Reese Richardson and Sholto David · Zenodo · https://zenodo.org/records/22090940 · Aug. 25, 2026 · DOI 10.5281/zenodo.22090940
2. At least 15 companies are selling antibodies using faked validation data · Reese Richardson · https://reeserichardson.blog/2026/08/25/at-least-15-companies-are-selling-antibodies-using-faked-validation-data/ · Aug. 25, 2026

Show 5 more sourcesHide sources

3. SF1 Monoclonal Antibody (OTI4C9), TrueMAB — TA809458 · Thermo Fisher Scientific · https://www.thermofisher.com/antibody/product/SF1-Antibody-clone-OTI4C9-Monoclonal/TA809458 · accessed Aug. 26, 2026
4. Primary Antibodies — Commitment to antibody specificity and reproducibility · Thermo Fisher Scientific · https://www.thermofisher.com/us/en/home/life-science/antibodies/primary-antibodies.html · accessed Aug. 26, 2026
5. Scaling of an antibody validation procedure enables quantification of antibody performance in major research applications · Ayoubi et al. · eLife · https://doi.org/10.7554/eLife.91645.2
6. Misleading Antibody Validation Images Controversy Expands to 15 Vendors · Shelby Bradford · The Scientist · https://www.the-scientist.com/misleading-antibody-validation-images-controversy-expands-to-15-vendors-74915 · Aug. 2026
7. Guidelines for Best Practices in Image Processing · Office of Research Integrity, US Department of Health and Human Services · https://ori.hhs.gov/education/products/RIandImages/guidelines/list.html · accessed Aug. 26, 2026

— Newsroom