Preprint servers and open publishing platforms
Preprint servers share manuscripts before formal peer review. Open publishing platforms can also make work available before review, while providing an explicit process for reviewing, revising and indexing it.
Server functions
A preprint is a complete scientific manuscript uploaded by its authors to a public server without formal review, often while also being submitted to a peer-reviewed journal. After a brief inspection for scientific content, readers can view it without charge; a 2022 account also describes posting without author fees and wider circulation before review.
Nature Precedings was described in 2008 as a free service for sharing presentations, posters, white papers, technical papers, supplementary findings and unreviewed manuscripts. It enabled rapid dissemination, comments, recording the provenance of ideas, archiving, sharing and citation before publication.
Rapid Research Notes, announced in 2009, offered open access to focused biomedical information of immediate interest to researchers, policymakers and the public, beginning with PLoS Currents: Influenza. Publishers rather than individual authors deposited material under National Library of Medicine guidelines; the material was not peer reviewed, but contributing publishers had to arrange expert-panel screening for acceptability.
Development in physics and biology
Early-1960s NIH-supported Information Exchange Groups circulated printed research memoranda by post for review and comment before publication, supplementing annual scientific meetings. Theoretical high-energy physicists were early users of electronic preprint sharing in the 1990s. ArXiv dates to August 1991 and covers physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics. A 2018 history places its initial development at Los Alamos National Laboratory and identifies Cornell as its supporter, whereas a 2019 account attributes its creation to Cornell. Its holdings exceeded 1.4 million preprints in 2018, 1.5 million e-prints by July 2019 and two million scholarly articles in the 2022 account.
Biological preprint exchange encountered fears about lost journal subscriptions, copyright infringement and unrefereed work circulating publicly. An NIH proposal during Harold Varmus’s leadership failed after the Federation of American Societies of Experimental Biology threatened to lobby Congress over NIH funding. BioRxiv, founded in 2013, was better received, possibly because open-access journals, arXiv’s established record and continuing publication delays had changed the circumstances. A February 2016 ASAPbio meeting explored broader preprint use in the life sciences; advocates saw opportunities for biology and medicine to follow the long-established physics model.
In 2018, bioRxiv was supported by Cold Spring Harbor Laboratory and the Chan Zuckerberg Initiative and accepted work across 26 biological fields. Neuroscience submissions had overtaken the early leaders, evolutionary biology and bioinformatics, by June 2018. By the end of September 2018, approximately 193,000 authors from 12,000 institutions in more than 100 countries had contributed nearly 33,500 manuscripts. Submissions exceeded 1,900 new papers monthly; reported traffic was 3.5 million page views and 750,000 PDF downloads, without a period specified for the traffic figures. The 2019 account expected bioRxiv to receive about 20,000 preprints annually.
The server’s cofounder attributed growth to journals’ willingness to consider preprinted manuscripts, user recommendations, readers finding good work, adoption at prolific institutions and scientists’ increasing Twitter use, with about 10,000 monthly tweets from the server account in 2018. Other servers included those hosted by the Center for Open Science.
Medical preprints and screening
In 2018, registered clinical trials could already be archived on bioRxiv, while medRxiv was being developed with the Yale Open Data Access project and other partners. Its intended purpose included making more trial data available, including results that journals commonly did not publish. The planned submission criteria covered funding, competing interests, trial registration, data sharing, reporting checklists, and ethics review and patient consent where applicable. The 2018 forecast anticipated a winter-2018 launch; creation was announced in June 2019, after controversial medical-server proposals dating from 2017.
The 2019 medRxiv service was managed by Cold Spring Harbor Laboratory, BMJ and Yale University. Each paper warned that it was unreviewed new medical research and should not guide clinical practice. Research articles were eligible, but case reports and opinion pieces were not. Screening considered potential harm to patients or the public; when in doubt, the server would decline posting and encourage publication only after peer review. Some journal editors opposed releasing medical manuscripts before review.
Open publishing and reproducibility
In the F1000Research model described in 2017, authors could publish conventional articles, incremental findings, methods, protocols, datasets and negative or null results rather than depend on an editor’s choice of narrative research papers. The platform required underlying source data to support reanalysis, replication and reuse, including software or equipment parameters and details needed to repeat the work. Its publisher presented wider author control, reduced delay and better reproducibility as advantages over conventional journals.
Platform selection should consider subject coverage, recognisable contributors or published work, affordability, open access and visibility to the relevant research community. In 2017, F1000Research’s publisher described it as one of very few, possibly the only, PubMed-indexed open science platforms.
Publication followed a pre-publication check for authorship criteria, scientific or medical relevance, legibility, absence of plagiarism, reporting standards and ethics; the work or substantial parts of it must not have appeared previously or be under consideration elsewhere. The publisher estimated that about 20% of submissions were rejected at this stage.
F1000Research review and indexing
After the checks, authors selected reviewers they considered appropriately qualified; the platform checked conflicts of interest and did not invite a reviewer without first asking the author. Typesetting and preparation of HTML, XML and PDF took an average of about seven days before online publication marked “awaiting peer review”, when formal review began. Reviewers disclosed their names and academic affiliations, and their reports appeared with the article.
The 2017 model used three reviewer statuses.
| Status | Display | Meaning |
|---|---|---|
| Approved | Green tick | No changes or only a few small changes requested |
| Approved with reservations | Green question mark | Significant changes requested to improve the paper |
| Not approved | Red cross | Very significant flaws and poor science; this status was described as rare |
PubMed indexing required either two approved reports or two approved-with-reservations reports plus one approved report. Authors could revise into new versions, with readers able to follow the initial paper, comments and revisions; the aim was a constructive, transparent conversation rather than hidden review.
Article assessment and funder platforms
F1000Research did not have or seek a journal impact factor in 2017, instead promoting article-level assessment through downloads, views, citations and Altmetrics coverage in social media, policy documents and news. Its publisher argued that open platforms could shift assessment towards the work’s intrinsic value, reduce barriers to data sharing, make negative and null results easier to publish, counter positive-results bias and reduce research waste and delay. An accompanying 2017 prediction was that easier publication of confirmatory, inconclusive and negative results would improve meta-analysis accuracy.
Wellcome Open Research used the F1000Research model to give the funder’s grantees a quicker, more transparent publication route with sufficient detail for reproducibility.