Open science is more than open access
The most significant change in research during the first two decades of the twenty-first century has been the rise of open science. This does not primarily mean explaining science to the general public. It means opening the research process to the scientific community as completely as possible, with transparency extending from the formation of an idea to the publication and subsequent use of the results.
There are at least two reasons this model has become increasingly necessary. One is the reproducibility crisis that spread across many fields during the past decade. Funding agencies and the public have become more concerned about whether research money is producing reliable knowledge, while scientists themselves have needed institutional mechanisms that make it possible to demonstrate that their work is trustworthy. The other reason is that technical progress and changes in the business model of academic publishing have made comprehensive transparency more practical. The economic and technological conditions are finally allowing some of these ideals to be implemented.
Scientists have always formed a kind of global, mildly utopian community. They obtain support from governments, private foundations, and companies, train younger researchers or collaborate with colleagues, produce new knowledge, and make it public through academic publication. Within this community, many resources are already relatively open. Large scientific facilities, for example, often accept applications from researchers around the world rather than restricting access by nationality. Applicants are screened for competence and generally pay their own travel and living expenses, while facility fees are often little more than symbolic compared with the actual cost.
Training grants and research funding are also, in principle, open to anyone who meets the requirements. The implicit barrier is usually education or verified scientific training; without such a barrier, the system would quickly be overwhelmed by unsupported claims. Across countries and regions, the terminology used in research is also unusually standardized. Someone who writes H2O does not need to share a native language with the reader for the meaning to be understood. Scientists have long tried to develop common languages and standards to reduce the friction of international communication, and English is now the main vehicle for that exchange.
Many ordinary forms of communication have similar histories. Email, for example, entered China partly through infrastructure established for research communication. International collaboration on questions in high-energy physics, climate change, and bioinformatics is so routine that it often requires no explanation. In practice, political considerations are more likely to obstruct scientific exchange than scientific ones.
What a fully open project would look like
A contemporary open-science project may have a complicated funding structure, but its presentation and outputs are generally expected to be as non-commercial and transparent as possible. Researchers may be encouraged to use open-source software instead of proprietary tools. Reusable results, such as viral sequences, should be submitted in standardized formats to databases commonly used by the relevant field.
A paper may be required to appear through open access, either immediately or after a fixed embargo. Figures and tables can be released under Creative Commons licenses, while the process used to generate them should be reproducible. A first draft may be posted to a preprint server and subjected to open peer review. During formal publication, the review history can also be made public so that readers can follow how the manuscript changed. After publication, the associated files, data, and processing scripts can be bundled and deposited in a public repository for others to inspect and reuse.
Meeting every one of these expectations remains difficult. The hardest parts are usually open peer review and providing enough detail for figures to be regenerated. Reviewers may raise questions that have not been fully resolved. If their comments are made public, the manuscript becomes vulnerable to a second round of criticism from readers who were not involved in the original review.
Reproducible visualization creates a different kind of problem. Many research graphics are made through graphical interfaces: a researcher clicks through a series of menus and settings, and even the person who created the image may not remember every operation later. Command-line scripts are much easier to reproduce, but they can impose a steep barrier on researchers who do not work in statistics or computing. Original data can be even more sensitive. Many research groups are reluctant to publish it openly and instead write in their papers that data are available from the corresponding author upon reasonable request.
Personally, I have never considered this level of openness unreasonable. Since my doctoral years, I have wanted to create a case that met the open-science standard from beginning to end. Earlier attempts failed for ordinary reasons. Some of the code needed to generate figures was tied to laboratory instruments and could not be separated from the equipment. In other cases, collaborators rejected open data release for various reasons and agreed only to provide data by email. Some journals considered ideas such as preprints and open peer review too unconventional and would not accept them.
Then came 2020, and I managed to complete such a case—although not without considerable detours.
Announcing the idea before the paper existed
To make the project more typical of an open-science experiment, and also to force myself to finish it, I adopted a strategy that might be described as making the promise first and earning it afterward. Beginning in 2019, I talked at conferences about developing mass-spectrometry-based “reactomics.” I had even invented the word while preparing a conference abstract. I later learned that the term had been used before, but not in connection with mass spectrometry in the way I intended.
By discussing it publicly, I made sure that people knew what I was trying to build and understood the basic principle before the paper was written. At that point, the only real question was how quickly I could produce a manuscript.
The writing itself was unexpectedly difficult. I knew the structure of a conventional research paper very well, but this was a data-mining and concept-introduction paper rather than a standard report of experiments. I had read relatively few papers of that kind. The concepts I knew had mostly come from textbooks, and textbooks do not explain how to turn a new concept into a publishable article.
After returning to China from a period abroad in 2019, I spent months thinking about how the manuscript should work. I finally produced a first draft near the end of that year. My supervisor had also never encountered a data-mining paper without experimental data in this form. His view was that it was not really a paper and should be split up and rewritten using a conventional structure.
I was convinced that doing so would remove the very feature that might give the work influence. I refused to change the central format, continued arguing for the original approach, and eventually obtained permission to place the draft on a preprint server. I then explained the idea in Chinese on a blog. During its first three months on bioRxiv, the manuscript remained among the ten most-viewed papers in the bioinformatics category, even though the work actually belonged to analytical chemistry.
The attention gave me some confidence, but the submission process was painful. I was not allowed to try a top journal because it was considered a waste of time. I eventually reached PNAS. After the manuscript passed between two editors, I received the view that the idea was good but the paper was too technical and its audience too narrow. The recommendation was to submit to a chemistry journal.
The obvious next attempt was JACS. The response was essentially the same: the work was too technical and should be redirected to another chemistry journal. After transferring it, I encountered a reviewer who appeared not to understand the subject and argued that a paper without experiments had no value. By then I was angry enough to set the manuscript aside, telling myself that I would return to it when I had time to convert it into another format.
The preprint, however, produced something much more useful than the earlier reviews. A research group in Chicago discussed the paper at a group meeting. Its principal investigator collected the comments and sent them to me. The message was, in effect: the paper has already been reviewed; now you decide how to revise it.
This was the first review I had received that contained substantive criticism. Earlier comments had not identified concrete problems, so responding to them would have meant guessing what the reviewers wanted. That kind of review does not improve a manuscript; it simply consumes time.
When I review other people's work, I have always treated the task as helping the authors make the paper stronger. Published manuscripts are not finished products, and a reviewer's role should be to identify ways to improve them rather than to display personal opinions. Even when I recommend rejection, I try to give the authors a direction for revision.
My own manuscripts have often received the opposite kind of treatment. Some reviewers could identify no problem at all, while others seemed determined to find a reason to reject the work. One earlier manuscript went through five or six rounds and was seen by more than ten reviewers. Each time, one reviewer would object without naming a specific problem. The paper was eventually rejected, and only later did I learn that some personal conflict unrelated to me had been involved. Such experiences are part of working in academic publishing, whether one likes it or not.
The route to publication
After responding publicly to the substantive open review, I prepared to submit again. The journal I planned to try was involved in a news controversy concerning discrimination against Chinese people the day before submission. I changed course overnight, revised the format once more, and sent the manuscript to Nature Communications.
More than half a month later, the response was that the idea was interesting but too technical, with a suggestion that I transfer the paper to one of the publisher's newer journals. The publisher had printed papers that were at least as technical, but for an author with no previous publication in the field and no prominent scientist lending the work credibility, the initial screening was especially strict. I also knew that if I refused the transfer, my supervisor would probably insist that I rewrite the work in a conventional format, so I accepted.
This time, a decision arrived within a week. I assumed the manuscript was being rejected again, but the reviewers had instead requested revisions. Their questions were not especially difficult. I was handling several manuscripts at once, so the paper sat with me for a few days. I sent the revision back, and it was accepted. The complete review-and-revision process took about a month.
The publisher's formatting and technical checks took much longer. The paper was accepted on October 2, but publication was delayed for more than a month. During that process, I noticed that the journal offered an option to publish the peer-review information. I selected it and released the review history as well.
Open review is hardly unusual in open-source software development. The same principle has also been used by low-cost, community-run publishing projects that are not yet conventional academic journals. Such models seem closer to the scientific utopia many researchers imagine: inexpensive, transparent, and centered on discussion rather than on commercial control.
A journal that used platforms such as GitHub or Gitee to provide low-cost open access and open review could make scientific communication more traceable. Open access has acquired a bad reputation in some circles because of predatory journals and excessive publication fees, but the underlying idea is not the problem.
Many journals founded and operated by scholarly associations already struggle financially. Some are eventually bought by publishing groups; others reduce their frequency from monthly to quarterly, then from quarterly to twice yearly, and gradually disappear. A community-based publishing experiment is therefore not guaranteed to survive. Still, if such a model could be made to work and then expanded, it might encourage conventional publishing to accept open review and post-publication review as normal parts of scholarly communication.
Publication should be a starting point rather than the end of a research conversation. New work almost always develops by modifying, testing, or correcting existing work. If those changes could be tracked publicly and the relevant versions updated, researchers would not need to write disconnected papers that repeatedly rebuild the same foundations.
One possible model would treat a scientific question more like software. Researchers working on the same topic could update a shared line of inquiry. Each meaningful contribution could produce a new version; if nobody continued the work, it would simply stop developing. A journal would define the topic, and the community would develop it collaboratively, much as contributors build a public knowledge base.
Such a system would need incentives and penalties, but those might be evaluated by the participants themselves. Question-and-answer communities have already developed useful ways of giving authors and reviewers transparent, traceable recognition. The results would not necessarily be efficient. De-commercialized publishing would almost certainly be slower and less polished. Yet if low-value papers were removed from consideration, the academic communication efficiency of many existing journals would not necessarily be much better. A considerable part of the current system mainly supports the publishing industry.
Releasing the data, code, and figures
For this project, I released the code and data. The repository even included the method used to retrieve and clean data from KEGG in real time. Every figure and table in the paper had code capable of reproducing it. I fixed the random-number seeds so that the output would remain consistent.
After the paper appeared, the publisher invited me to write a blog post explaining the project. That was easy enough; I had already described it in Chinese a year earlier. At that point, the case looked like a nearly complete example of open science.
It was not complete yet.
The concept of reactomics had been constructed by me from beginning to end, so I decided to take the experiment one step further. I began putting the project's ongoing progress into an online slide deck. The slides could be reused in talks and updated continuously as the work developed.
This was a much riskier form of openness. I was willing to publish the entire progress deck only because the project did not use data from my current laboratory. All of the evidence came from publicly available data. That distinction mattered. A concept can often be validated through carefully controlled experiments designed by the person who proposed it. Testing it with someone else's data is more demanding: the concept must be sufficiently general and robust to work in an experiment that was never designed for it.
In this case, I used urine samples from patients with lung cancer. That was also the source of my confidence. If an idea or conceptual framework is genuinely sound, its predictive power should be strong enough to survive data collected for another purpose.
Publishing the project's progress in real time—including material that had not yet been published—felt like writing a fair-play mystery novel. The reader could follow the clues and try to discover the answer alongside the researcher. This may be one of the most radical forms of open science: not merely releasing the finished paper, but exposing the investigation while the problem is still being solved.
The process was genuinely exciting. I knew that several research groups were using the tools and ideas I had developed. Some contacted me, while others did not. I chose not to distinguish between the two. Anyone who wanted to look could look. The project was being played with the cards face up. I was not short of ideas; what I feared was that good ideas would receive no follow-up at all.
Recordings of my conference presentations were also made public. I did not worry about someone pointing out that I was wrong. Errors are part of how I improve. My real goal was to use available technical tools to distribute research results more effectively.
I used several public channels, including a Chinese social-media account, a question-and-answer column, and an English blog hosted by the publisher, to demonstrate that this kind of open-science workflow could function in practice. The paper had only two authors. A more aggressive submission and disclosure strategy would probably have been difficult to coordinate in a larger team. In reality, the manuscript was discussed mainly by my supervisor and me, and my supervisor was more often the person urging me to abandon the project than the person encouraging it. Two people working together can already have negative combat power; adding more collaborators might have turned the coordination into a disaster. It is also possible that I simply lacked the ability to manage a larger team.
A practical test rather than a slogan
The paper was not published in an especially prestigious journal. That was not the main point for me. The project gave me a complete experience of open science, so future discussions would no longer be based only on theory.
The preprint received more than 700 full-text views, and the paper received close to 500 views during its first week after publication. I was satisfied with that response. Most of my other papers had been downloaded fewer times than a small fraction of that number.
The experience has made me more willing to use open-science practices in future research. Research should not be carried out in fear of being tested. My work is in basic research, and there is usually a considerable distance between a new result and an application or a profit-making product. If the results cannot be communicated effectively, there is little value in guarding a small private territory and waiting for recognition.
The more useful attitude is to make the work inspectable: disclose the idea, release the data when possible, show how the figures were made, publish the criticism, and allow others to test the claims with their own tools and evidence. A research paper can then become not a sealed container of conclusions, but a visible stage in a continuing process.