Fotis Psomopoulos1
1 Institute of Applied Biosciences (INAB), Centre for Research and Technology Hellas (CERTH), Thessaloniki, Greece
fpsom [at] certh.gr
Abstract
As a result of many years of global efforts, ensuring that data are Findable, Accessible, Interoperable and Reusable (also known as FAIR) is nowadays a clear expectation across all science domains. While data and data management have been the primary focus across many activities, research software has only recently started getting similar attention. As a result, a coordinated effort by the wider community allowed to redefine and extend the FAIR principles to research software, with similar activities now in progress aiming to enhance reproducibility, quality assurance, and long-term sustainability in software development.
At the same time, we see the emergence of the field of artificial intelligence (AI) and machine learning (ML) as a key technology impacting all sciences. As AI algorithms and models become increasingly integrated into scientific workflows, there is an urgent need to maintain high standards for research software, with the reliability and quality of the underlying software being of primary concern.
High-quality research software is definitely a key catalyst in that direction. In this context, “quality” involves not only creating robust and efficient algorithms, but also implementing rigorous quality control processes throughout the software lifecycle. There are several initiatives (such as ReSA, Turing Way and SSI) that are making available best practices, guidelines and recommendations on research software, from design and coding to testing and deployment, as well as major funded projects (such as EVERSE).
Another key aspect is around benchmarking, as it serves as a critical tool for evaluating performance, scalability, and generalizability of AI solutions across diverse datasets and use cases. In order to effectively run a benchmarking process, it is essential to establish standardized benchmarks and evaluations protocols, as well as the respective underlying services and infrastructure to facilitate this. In both cases, input and direct involvement of the respective community is essential, in order to fostering transparency and comparability in AI research.
Finally, beyond the technical aspects, there is a clear need for a coherent effort towards the interpretation of the actual FAIR principles for ML. Some efforts already exist, such as the RDA FAIR4ML interest group, as well as the efforts under the AI4EOSC project and the ELIXIR infrastructure. However, we still have some way to go, and direct community involvement is critical to ensure both wide adoption and ultimately uptake of these practices.
Keywords: bioinformatics, research software, machine learning, FAIR principles, software quality
Acknowledgement: This work has been written with the support of the EVERSE project, funded by the Horizon Europe Framework Programme (HORIZON-INFRA-2023-EOSC-01- 02) under grant agreement number 101129744. The views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Executive Agency (REA). Neither the European Union nor the granting authority can be held responsible for them. This work was also supported by ELIXIR, the research infrastructure for life science data.