Problems with the Current Approach to the Dissemination of Computational Science Research and Its Implications for Research Integrity

Computational methods are at the heart of all twenty-first century research, but the acceleration of the application of computational approaches to biomedical research is particularly striking. From the simulation of the behaviour of complex systems, through the design and automation of laboratory experiments, to the analysis of both smalland large-scale data, computational approaches, and the well-engineered software that underpins them, have proved to be capable of transforming biomedical research. In parallel, the growth of high-throughput technologies and continual innovation in hardware, imaging, sensing and monitoring, demands unprecedented levels of collaboration between computational and experimental scientists, to continue the transformation of biology andmedicine fromprimarily descriptive to quantitative, predictive disciplines. As a result, biomedical research is dependent as never before on computational science methods and hence also on the instantiation of those methods in research software.1 Despite this central and rapidly growing importance, biomedical research scientists are rarely trained to develop their own well-engineered software, nor are they trained to understand what it takes for research software to be transformative. Instead, research software is typically developedwith the primary goal of facilitating rapid publication of a research group’s most recent results in the scientific literature. It is not usually made available to the research community, or even to reviewers, and so is not (and cannot be) verified.2 Significant research time is lost (usually by Ph.D. students with no formal training in software development) in re-implementing already-existing software tools from (often inadequate) literature descriptions. Even if successful, the