C'mon! Time to Change How We Share Research Products

Ask any middle schooler to find a song, new or old, and they will be able to play it for you in seconds. Ask a graduate student with years of specialized training to find the data associated with a publication for further analysis, and, after months of manual work, the result will likely be disappointing. I think the analogy is valid; globally, millions of papers and tens of millions of songs are published every year. Over centuries, musicians and their industry have adopted new technology to share the product of their work, lowering access barriers. Yet scientific publications have barely evolved since the first paper was published in Philosophical Transactions in 1665 [1], see Figure 1. While searching for papers is easier now than when I was a student in the 1990s, accessing the underlying data and procedures remains challenging. The raw outcome of the day-to-day work of researchers in the physical sciences and engineering is data and procedures, which are then turned into knowledge and new technologies by confirming or denying hypotheses, or developing new materials, molecules, and devices. Papers and reports, as the main avenue for the dissemination of these outcomes, are less than ideal for two main reasons. First, natural language, figures, and printed tables are not the best media to share data and procedures. Data belong in findable, accessible, interoperable, and reusable (FAIR) repositories or databases, and procedures should follow domain-specific formats and best practices. Unfortunately, most communities lack formal processes and standardized formats to share research procedures, which hinders reproducibility. One example of recent progress is Sim2L [2], which provides a mechanism to document, publish, and share computational workflows. Second, only a biased subset of the research data generated is shared in publications and reports. This is problematic for human readers and very limiting for artificial intelligence (AI) models. By keeping most raw data inaccessible outside of small groups, we limit their impact. Only the interpretation of the original authors is available, and we preclude the data re-use (maybe in aggregation) to generate new knowledge. The potential impact of sharing data in meaningful ways can be estimated by assessing select efforts that have made data availability following FAIR principles [3] a priority. For example, the Materials Project [4] contains >600,000 density functional theory calculations accessible via a web interface and API. Over 30,000 publications have cited the project, and its data have sometimes been used in ways the developers did not foresee [5]. Other platforms are having a similar impact and slowly changing how research is performed [6-9]. Funding agencies and governments have enacted data share mandates, and most publishers require data share statements. These are important but insufficient steps. Continuing with the analogy between research data and entertainment, focusing on data sharing at the time of publication is the equivalent of Netflix not collecting usage data and then asking customers once a month to go back and list the movies and shows they watched. The common practice of thinking about data availability at the time of publication is wasteful in terms of resources and takes the researcher's attention from the task at hand, which should be communicating their findings to others. In my experience, there are relatively simple steps research groups and communities can take to improve data practices. A recent publication discusses specifics and available resources [10]. Before a research effort is launched, it pays off to develop a data plan that spans the entire data lifecycle, from its generation to its final publication and permanent storage. It should include intermediate steps like turning raw data into FAIR data, connecting them to AI/ML, analysis, or visualization tools. I would like to highlight two aspects that are often misunderstood. First, procedures, both experimental, theoretical, and data analysis, should be documented with appropriate tools and shared following FAIR principles. We currently have several tools to document, share, and publish procedures and data analysis workflows [2, 11]. I believe the wide adoption of these tools will accelerate scientific and research computing. Second, FAIR principles do not mean unrestricted, open, or free access. While using FAIR principles to broadly share results is expected to accelerate progress and should be our goal, there are many legitimate reasons to limit access or include embargo periods. FAIR data principles are still valuable even when data access is restricted to an organization, group, or institution. I believe organizations and communities that accelerate this transition will experience significant benefits. This would be particularly beneficial to specialized research fields, like energetic materials, where coordination can be straightforward and data aggregation is critical. C'mon!