A blood test for Alzheimer’s and Parkinson’s is one step closer, thanks to two new datasets designed to train artificial intelligence on problems it has never been able to tackle before. The first dataset, EV-Bench, focuses on tiny particles called extracellular vesicles that circulate in the blood. These particles carry RNA from the brain, but chemical modifications to that RNA—key markers of neurodegenerative and psychiatric disorders—have been nearly impossible to study without invasive procedures. EV-Bench combines natural and synthetic RNA to give AI a training ground for detecting those modifications from a simple blood sample. The second dataset, SPDE-Bench, targets the equations that describe turbulent weather and ocean currents, providing standardised data so AI can help solve them. If successful, these benchmarks could transform diagnostics for brain diseases, replacing spinal taps or brain scans with routine blood tests. They could also accelerate climate modelling by giving AI the structured data it needs to predict storms and currents. The project is fundamentally about building infrastructure—creating the shared, open-source tools that allow researchers across disciplines to compare methods and collaborate. By March 2026, the team will release initial versions of both datasets, along with documented AI models and protocols.
View original technical description
This project aims to create two world-first benchmark datasets to accelerate the use of artificial intelligence (AI) in science. These datasets—EV-Bench and SPDE-Bench—are designed to unlock AI’s potential in two critically important and scientifically complex areas: brain health and climate-related physics. EV-Bench will enable AI to detect chemical modifications in RNA found in brain-derived extracellular vesicles (EVs)—tiny particles circulating in the blood. These RNA modifications are crucial to understanding neurodegenerative and psychiatric disorders, but until now, they have been extremely difficult to study without invasive procedures. By creating a high-quality dataset that combines native and synthetic (in vitro transcribed) RNA, EV-Bench will allow researchers to explore brain health through a simple blood test, opening the door to new diagnostics for diseases like Alzheimer’s and Parkinson’s. SPDE-Bench targets physical sciences, providing a structured dataset for solving stochastic partial differential equations (SPDEs)—the equations that describe turbulent weather, ocean currents, and other dynamic systems. Although theoretical breakthroughs have advanced our understanding of SPDEs, applying AI to solve them has been limited by the lack of standardised data. SPDE-Bench fills this gap with reproducible, open-source data and baseline AI models, allowing researchers to compare methods and advance the field collaboratively. The project will also foster interdisciplinary partnerships. By bringing together mathematicians, clinicians, AI researchers, and domain scientists through workshops and hackathons, it will promote collaboration and ensure real-world adoption. Partners include Imperial College London, the Rosalind Franklin Institute, the Institute of Molecular and Computational Medicine, and the European Centre for Medium-Range Weather Forecasts. By March 2026, the team will deliver initial versions of both datasets, alongside documented AI models and protocols for scientific use. These benchmarks will support the UK's ambition to lead in trustworthy, high-impact AI and create infrastructure that underpins future investments across clinical and environmental research.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know