henhao
Sign in
Scientific Communication & EducationbioRxiv

Tracking the popularity and outcomes of all bioRxiv preprints

R. J. Abdill, R. Blekhman

Abstract

Researchers in the life sciences are posting their work to preprint servers at an unprecedented and increasing rate, sharing papers online before (or instead of) publication in peer-reviewed journals. Though the popularity and practical benefits of preprints are driving policy changes at journals and funding organizations, there is little bibliometric data available to measure trends in their usage. Here, we collected and analyzed data on all 37,648 preprints that were uploaded to bioRxiv.org, the largest biology-focused preprint server, in its first five years. We find that preprints on bioRxiv are being read more than ever before (1.1 million downloads in October 2018 alone) and that the rate of preprints being posted has increased to a recent high of more than 2,100 per month. We also find that two-thirds of bioRxiv preprints posted in 2016 or earlier were later published in peer-reviewed journals, and that the majority of published preprints appeared in a journal less than six months after being posted. We evaluate which journals have published the most preprints, and find that preprints with more downloads are likely to be published in journals with a higher impact factor. Lastly, we developed Rxivist.org, a website for downloading and interacting programmatically with indexed metadata on bioRxiv preprints.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (2 Apr 2019) · CC BY 4.0

Crawled metadata for all 37,648 bioRxiv preprints through November 2018 reveal accelerating submissions and downloads, 67.0 percent publication of 2013-2016 preprints, and a public Rxivist database.

Preprints are changing biology publishing and funder policy, yet systematic data on bioRxiv use (which fields post most, how heavily preprints are read, how many reach journals and how fast) were scarce. The study answers these questions for the server's first five years and releases Rxivist, an API, website and database dumps that replace the ad hoc scraping earlier bibliometric studies needed.

Key results

  1. ShownbioRxiv submissions are accelerating: output in 2018 alone surpassed the previous four years together, and neuroscience is both the largest collection and the main driver of growth.

    37,648 preprints by end of November 2018; 18,825 posted in the first 11 months of 2018; October 2018 was the first month above 2,000 postings (2,119, a 6.3 percent rise); neuroscience reached 6,000 preprints first, ahead of bioinformatics (4,249) and evolutionary biology (2,934). · Preprint submissions

  2. ShownReadership is growing fast. Downloads peak in a preprint's first month and then decay. Neuroscience leads total downloads because it has more preprints, not because each one is read more.

    1,140,296 downloads in October 2018, 82 percent above October 2017; 19,699,115 downloads from November 2013 to November 2018; site-wide median 279 per preprint (genomics highest, 496); neuroscience median 269.5 vs 281 for other categories (Mann-Whitney U p = 0.0003); monthly medians 73, 46, 27 for months one to three. · Preprint downloads

  3. ShownMost older preprints eventually appear in journals. bioRxiv's automated linking to published versions seems to miss many of them, especially in earlier years, so official publication rates are likely too low.

    15,797 preprints (42.0 percent) linked to 1,531 journals; 67.0 percent of 2013-2016 preprints vs 20.0 percent of 2018 preprints published; manual search of 120 preprints flagged unpublished found 37.5% already in journals (53 percent for 2015, under 17 percent for 2017); all 40 sampled published links were correct. · Publication outcomes; Estimation of ranges for true publication rates

  4. ShownPreprints that land in higher-impact journals have more bioRxiv downloads, and published preprints are downloaded about twice as often as unpublished ones. Which causes which is not established.

    Journal-level Kendall's tau-b 0.5862 (p = 1.364e-06); preprint-level tau-b 0.2053 (n = 7,445, p = 9.311e-152); Nature Methods (2017 JIF 26.919) median 2,266 downloads vs PLOS ONE (2017 JIF 2.766) median 279; published vs unpublished medians 394 vs 208 (Mann-Whitney U p < 2.2e-16). · Publication outcomes

  5. ShownThe median time from posting on bioRxiv to journal publication is about 5.5 months. That is longer than bioRxiv's own 2016 figure, and the delay differs significantly between journals.

    Median 166 days; 75 percent within 247 days, 90 percent within 346 days; 23.9 percent above the 134-day median bioRxiv reported in mid-2016; among the top 30 journals, Nature Genetics slowest (272 days) and G3 fastest (119 days) (Kruskal-Wallis p < 2.2e-16; Dunn's q < 0.05, Benjamini-Hochberg). · Publication outcomes; Retrieval of publication date information; Calculation of publication intervals

Methods, data and reproducibility

A Python 3 crawler (Requests-HTML, psycopg2) scraped metadata, monthly downloads and journal links for every bioRxiv preprint into PostgreSQL, adding Crossref publication dates; R analyses used Mann-Whitney U, Kendall rank correlation, Kruskal-Wallis and Dunn's tests, with manual Google Scholar checks of publication linking.

  • Metadata, monthly download counts and journal links for every bioRxiv preprint posted November 2013 to November 2018 across 27 subject categories · n = 37,648 preprints · 10.5281/zenodo.2465689 (database snapshot and figure data)
  • bioRxiv author records (names, affiliations, emails, ORCID identifiers) consolidated by exact name or ORCID · n = 170,287 authors
  • Published preprints with journal names and Crossref publication dates · n = 15,797 preprints in 1,531 journals
  • Random sample of preprints not marked published, manually searched via Google and Google Scholar (30 per year, 2014-2017) · n = 120 preprints
  • Sample of preprints marked published, checked to assess linking specificity · n = 40 preprints (10 per year)
  • Published preprints with download totals and journal impact factor for preprint-level correlation · n = 7,445 preprints
  • Web of Science article counts and 2017 journal impact factors for the journals publishing the most preprints · n = 30 journals
  • Stop Predatory Journals list screened against preprint publication venues · n = 1,345 journals
Methods & tools
Rxivist (website, API, database), Requests-HTML, psycopg2, PostgreSQL, Docker, Crossref Metadata Delivery API, R (stats package), Mann-Whitney U test, Kendall rank correlation (tau-b), Kruskal-Wallis and Dunn's tests with Benjamini-Hochberg correction, Web of Science, Google Scholar
Code
https://github.com/blekhmanlab/rxivisthttps://github.com/blekhmanlab/rxivist_webhttps://github.com/blekhmanlab/rxivist/blob/master/paper/figures.md
Data & accessions
https://doi.org/10.5281/zenodo.2465689https://rxivist.org/https://www.rxivist.org/docs
Software & environment
Python 3 crawler using Requests-HTML and psycopg2; PostgreSQL database; API, crawler and website each run in separate Docker containers (Docker is the sole dependency); figures generated with SQL queries and R code in figures.md.
Experimental conditions
Loading rankings through psql copy cut the ranking step from several hours to less than one minute; no hardware specifications reported.
Reproducibility gaps
Software versions are not pinned. The live Rxivist data are re-crawled every two to three weeks, so only the Zenodo snapshot reproduces the November 2018 figures.
Limitations (6)
  • Publication status comes from bioRxiv's linking, which appears to rely on title matching and misses published preprints (37.5% of a 120-preprint manual sample). Year-level corrections carry a 17.89 percentage-point margin of error. · stated by the authors
  • The link between journal impact factor and downloads is only a correlation. Downloads were not timed relative to journal publication, so appearing in a journal could drive later downloads rather than the reverse. · stated by the authors
  • Authors are matched only by exact name or ORCID, and only 17.4 percent have an ORCID. This produces duplicate authors and merged namesakes (e.g., 'Wei Wang' with 40 preprints). Institution counts are skewed by inconsistent affiliation names. · stated by the authors
  • Publication dates come from Crossref metadata that was not validated, and publishers define those dates differently, which may inflate posting-to-publication intervals. 66 preprints dated before their bioRxiv posting were excluded. · stated by the authors
  • The journal-level correlations and category comparisons do not adjust for posting year or field. Both strongly affect download totals, so the impact factor association could be partly confounded by preprint age and subject mix. · noted in this summary
  • Downloads come directly from bioRxiv's counts, and no filtering of automated or repeated traffic is described, so download totals may not reflect actual human readership. · noted in this summary

Who should read:Read this if you study preprint adoption and scholarly communication in biology, or want programmatic access to bioRxiv metadata (downloads, publication links, authors) through the Rxivist API and database dumps.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISFraser '19Fu '19Sever '26Fraser '21Satish '20Fraser '20
82% semantic matchScientific Communication & Education
N. Fraser, F. Momeni, P. Mayr+1 more·bioRxiv·Jun 22, 2019