The Materials Research Platform: Defining the Requirements from User Stories

Abstract

A recent meeting focused on accelerated materials design and discovery examined user requirements for a general, collaborative, integrative, and on-demand materials research platform. A recent meeting focused on accelerated materials design and discovery examined user requirements for a general, collaborative, integrative, and on-demand materials research platform. What common elements are necessary to create a general framework for materials innovation? Here, we provide a retrospective analysis of high-level themes that emerged from a focused discussion on the requirements for a future materials research platform. These discussions occurred during the annual Toyota Research Institute—Accelerated Materials Design and Discovery meeting (May 29, 2019, in Boston, MA) with more than 40 field experts from universities, US national laboratories, Toyota Research Institute in the United States, and Toyota Motor Corporation in Japan. These researchers contributed ideas toward what capabilities such a platform should have to accelerate the design and discovery of materials. To maximize the information captured from these field experts' ideas, we followed a strategy comprised of exploitation and exploration inspired by knowledge acquisition strategies in machine learning. In a first session, we matched researchers with themes that we expect them to be knowledgeable about (hence exploitation). In a follow-up session, we did a quasi-random assignment of researchers to themes (hence exploration), with the goal of capturing unique ideas that might elude experts embedded deep in their work. Sixteen pre-selected themes relevant for a materials research platform were discussed twice in two unique groups, one formed as an exploitation and the other formed as an exploration team. The selected themes were: Adaptive systems—active-learning and beyond; Automation of experiments; Automation of simulations; Collaboration; Data ingestion and sharing; Integration; Knowledge discovery; Machine learning for experiments; Machine learning for simulations; Multi-fidelity and uncertainty quantification; Reproducibility and provenance; Scale bridging; Simulation tools; Software infrastructure; Text mining and natural language processing; and Visualization. We digitally recorded the ideas in the form of “user stories” from agile software development practices. The meeting captured many stories that would be considered obvious, but several unexpected ideas emerged. An initial parsing revealed three interrelated themes for the design of a useful platform: (i) data and knowledge assets, (ii) automation of science, and (iii) integrative approaches, as outlined in Figure 1. The ideas related to data and knowledge assets rely on the FAIR principles1Wilkinson M.D. Dumontier M. Aalbersberg I.J. Appleton G. Axton M. Baak A. et al.The FAIR Guiding Principles for scientific data management and stewardship.Sci. Data. 2016; 3: 160018https://doi.org/10.1038/sdata.2016.18Crossref PubMed Scopus (5366) Google Scholar but with added distinguishing capabilities relevant for a materials research platform. For example, the platform should enable sharing and collaboration—not just around data, but also knowledge assets such as machine-learning models or scientific workflows. Baselines to gauge new findings are critical but often overlooked. Artificial intelligence (AI)-assisted search and visualization could amplify the scientific abilities of human researchers. As part of automation of science, experimental workflows are envisioned to be run “on-demand”, where tasks are picked up by relevant laboratories with fully or partially automated experimental capabilities, forming collaborative networks via the platform. Given the state-of-the-art for automation in modeling and simulation, a similar but more “productized” automated capability with web-based user interfaces is envisioned to assist researchers to run on-demand simulations complementary to experiments, or train or use on-deman