Modeling Convergence Undercurrents in Brain Science via Statistics and Machine Learning



Journal Title

Journal ISSN

Volume Title



From Leonardo da Vinci to the Nobel laureates, the science exemplar changed from the polymath to the dedicated specialist in response to evolutionary pressures. It is now believed that we are in the early stages of yet another evolutionary adaptation, where specialists from different disciplines need to collaborate to solve complex multidisciplinary problems. This idealized integration process, known as convergence, emerges as the new science exemplar. Despite the consequential nature of such a paradigm shift, the exact operationalization and efficacy of convergence remain unclear. To provide much needed answers to these two questions, we identified brain science as a unique convergence testbed to base our study. By jointly analyzing the disciplinary pedigree of the authors with the subject areas of nearly one million brain science publications between 1980 and 2019, we exposed a seeping convergence undercurrent: Science integration does not only neatly take place among researchers from different disciplines, but also awkwardly within researchers through expansive learning. Our models reveal three key findings: First, brain researchers tend to tackle subject areas beyond their core expertise, especially when these areas are epistemically close - a convergence shortcut. Second, this expansive learning behavior appears to precipitate and compete with true convergence, although it is clearly less impactful. In a finite science ecosystem, transnational epistemic moves by individual researchers effectively crowd out collaborations among distant disciplines, from where breakthroughs to grand challenges usually emanate. Third, major funding initiatives in brain science unknowingly promote shortcuts to convergence research. Regarding the content of the research publications in our corpus, in addition to linear model analysis based on Medical Subject Headings (MeSH) keywords, we also applied Machine Learning (ML) analysis on the articles’ abstracts. The ML methods yielded results that either validated or complemented those of the linear models. Furthermore, ML furnished insights regarding the timing and source of transformative developments in brain science that elucidate the abstract conclusions of the linear models. Such insights include the role and effect of Magnetic Resonance (MR) imaging and data analytic methods in brain science advancements.



Convergence, Cross-disciplinary, Machine Learning, NLP, Panel Modeling