An algorithm to police figure duplication?

Inappropriate figure duplication in publications is a surprisingly prevalent form of scientific misconduct. Perhaps the most infamous example in recent years was the fraudulent duplication of figure regions in the ‘STAP (stimulus-triggered acquisition of pluripotency)’ cell paper, a story that made headlines worldwide and contributed to the retraction of the paper in question. But how can such malpractice be effectively policed? Some journals manually screen images in submitted manuscripts — a laborious and time-consuming task. However, this process could potentially be automated.

A new study, published on the bioRxiv pre-print server, uses an algorithm to seek out duplication of figure regions, even after manipulation. The authors, Acuda et al, analysed 2 million figures from 760,000 open-access articles. Potential instances of duplication that were identified by the algorithm and machine learning were then reviewed by an author panel. The authors estimated that 9% of figure duplication was ‘suspicious’, while 0.6% could be considered fraudulent. Crucially, nearly half (43%) of inappropriate figure re-use occurred across articles.

Such technology could offer a more streamlined, rapid and accurate approach to figure screening by journals and aid scientific integrity. Publishers, however, would need to ensure a unified approach to successfully eliminate figure duplication across the literature.

——————————————————–

Summary by Emma Prest PhD from Aspire Scientific