QED Science published a list of 574 biology preprints selected by its AI system
QED Science published a list of 574 biology preprints selected by its AI system
On August 10, Nature reported on QED Science. The company says that its system evaluated 57 455 bioRxiv preprints published between May 2025 and April 2026 without access to author names, institutional affiliations, or journal status. Of these, 574 were included in The 1% list. In November 2025, bioRxiv and QED launched the first pilot of an external service for authors.
A preprint is a manuscript that researchers make publicly available before formal peer review by a journal. At that stage, the work is available to read but has not yet acquired journal status. QED adds automated analysis to this stage. According to the company, the service identifies the authors’ claims, checks whether the presented data support them, and evaluates the study design, statistical methods, and reporting of results.
After submitting a preprint, an author can send it to QED through their bioRxiv account. bioRxiv reports that the review usually arrives within about half an hour. It identifies which conclusions are supported by the data and where the author may need to revise the text or analysis.
QED cofounder Niv Mastboim explained the aim to Nature as follows:
“The goal was to judge the science by the work itself, rather than by the journal or the prestige of its authors.”
On June 24, QED published The 1%, a list of 574 preprints that received the highest scores from its system within this corpus. The author report provides feedback on a specific manuscript. The 1% list turns the same score into a public recommendation for other researchers.
In the DDD Benchmark, an evaluation of AI models, the developer chooses between a private report and a shared table. DDD compares models, while QED evaluates individual manuscripts. In both cases, the published assessment becomes a reference point for people who did not participate in the evaluation.
In its own white paper, QED describes how it validated its score. The company compared the QED Score with SJR, a citation-based measure of journal reputation. QED selected 100 pairs of preprints for which the two rankings differed most, recruited 15 subject matter experts, and concealed the journal names and QED scores from them. In the 60 comparisons where the experts preferred one of the two papers, their choice agreed with the higher QED score in 75% of cases.
In the bioRxiv pilot, the service gives authors an early assessment of their manuscripts. The 1% list turns this score into a public recommendation and may influence which preprints other researchers choose to open first.