An EPSRC National Research Facility to facilitate Data Science in the Physical Sciences: The Physical Sciences Data science Service (PSDS)

The modern physical scientist cannot perform their research without generating significant quantities of data, having recourse to related/prior data, significant data analysis and integrating results with other data. This requires a range of skills and resources that are not available to the majority of physical scientists. There is therefore an urgent need in the physical sciences for providing access to data and integrating them with data science approaches. This requires building a new skills base that enables and empowers working in a data science way.

The Physical Sciences Data-science Service (PSDS) will provide a single place where existing databases, open data sources and data that is still being worked on can be stored and searched in a unified way. This means that it will become trivial to find and combine different types of physical sciences data - from details on structure to measured physical properties of materials. It will also make possible instant comparison of and context for experiment data with that already available.

This is just the start however. There is enormous potential for being able to perform data science across all of these data, that is for example, Machine Learning and Artificial Intelligence approaches, which are becoming a new avenue of research in their own right.

It is vital that data science becomes a routine tool for all physical scientists. For many this will mean learning new skills. The PSDS will therefore develop a training programme around the four main competencies (statistics, programming/tools, computational methods & data visualisation) required to perform data science. Identified links with networks and postgraduate training will enable PSDS users to gain deeper skills in various aspects of data science.

The long-term aim is for the PSDS, and therefore data science, to become a seamless, key part of the research infrastructure for physical scientists.

The purpose of the service is to facilitate new data science approaches to research for a broad cross-section of the physical sciences research community. The impact on the communities and disciplines within physical sciences will be very significant. By providing data resources and analysis processes that would not otherwise be available to the researcher, entirely new results or avenues of research will open up. This will be further enhanced due to the way in which it will be possible to blend and analyse data across different resources. This is not just a matter of breaking down silos, but rather fuelling far-reaching research. The vision is that large aggregations of data can be generated and systematically analysed using machine learning or artificial intelligence approaches. This would effect a paradigm shift in the way in which physical sciences discovery and analysis is performed, so resulting in new science that quite simply could not be achieved by traditional methods.

Early career researchers will be significantly impacted by this service. Our training programme will be particularly aimed at this audience, taking those with a sound, but traditional, physical science background and exposing them to new, emergent methods that promise to revolutionise the subject area. It is the earlier generation of researchers that are best poised to take advantage of this new approach and our training programme is founded in a data science training framework designed specifically to develop researchers in this new direction - we will provide the first rungs on the ladder and link into networks and training schemes that can develop trainees further. We expect a new type of researcher to result from this approach.

The providers of data resources will be particularly impacted. In an open environment where new aggregations of data can be developed on the fly depending on a specific research question, there is enormous opportunity. Data that is currently siloed and therefore only useful to narrow communities will now be available for exploitation in a multitude of new ways that were previously unimagined. The ability for these new collections of data to be mined or systematically analysed provides further opportunity for data providers as well as users.

These new approaches that liberate data and make it available for large scale systematic analysis will ultimately benefit the materials science and chemical manufacturing industries. Time to discovery will be reduced, optimisation and efficiency in manufacture enabled, and new and more elaborate property-driven products engineered. The UK high-value industry needs many more data-science minded researchers and this service will provide a 'low barrier to entry' method for a large number of traditionally educated researchers to engage.