Dementia is a progressive, chronic condition with no cure, and its prevalence is rising, with experts predicting the number of people affected will almost double by 2050.
But hope for new treatments targeting dementia and other non-curable diseases is growing, thanks to a wave of European startups applying AI to scientific discovery.
Prima Mente, an AI neuroscience company based in London, is leading a study into dementia, across 1,000 patients in 15 healthcare settings. It combines blood biomarkers and genotyping with remote cognitive testing to enable earlier, more accurate detection of cognitive decline.
Hannah Madan, Prima Mente’s cofounder, says the startup is trying to imagine a world where Alzheimer’s, dementia and other neurodegenerative conditions are not treated as one big bucket of disease, but broken down molecularly — a little like cancer, which now has specific treatments for specific types.
“It’s a simple study, designed to bring innovation to the neurocare pathway,” says Hannah Madan, Prima Mente’s cofounder. “The NHS teams we’re working with are super excited.”
The potential for AI to revolutionise the healthcare sector is no longer theoretical. Researchers, doctors and clinicians are using models to analyse medical images and diagnose conditions earlier; to design and run more successful clinical trials faster; and to map the right therapies to the right patients with a higher degree of accuracy.
In 2024, two Nobel Prizes were awarded to scientists using AI to shape the future of medicine. In the future, experts believe it’ll be possible for doctors to create digital twins for each patient so they can simulate treatment plans and transform personalised healthcare.
It’s an astonishing pace of change, but there are still logistical challenges preventing teams from taking full advantage of this technology.
Infrastructure challenges
To facilitate their important work, the team of neuroscientists and AI researchers behind Prima Mente has spent the past year developing Pleiades. It’s the first epigenetic foundation model that can spot signs of early neurological disease with a high degree of accuracy. But it requires an enormous amount of elastic, cost-efficient high performance computing (HPC).
The infrastructure many teams are working with just isn’t built for modern research.
“We are pre-training huge models in-house, with up to 80bn parameters,” says Madan. “That infrastructure is something only Uber, Wayve, Google DeepMind and other big tech companies used to have access to.”
The biotech startup is now leveraging Nebius’s infrastructure to power its research. Nebius is the first Nvidia Reference Platform cloud partner HQ’d in Europe, and it recently announced the deployment of Nvidia Blackwell Ultra GPUs in the UK. It’s now one of the largest independent AI infrastructure builders globally, with data centres in six countries — a resource that healthcare innovators desperately need. In June, Nebius held its first AI Discovery Awards for biotech startups, awarding nearly $1 million in Nebius AI Cloud credits. Four top teams received $100,000 each in GPU credits, with additional prizes going to runners-up and honourable mentions.
“The infrastructure many teams are working with just isn’t built for modern research,” says Dr Ilya Burkov, global head of healthcare and life sciences growth at Nebius. “It’s clunky, expensive to maintain and often isn’t easily scalable. When teams can’t access the compute power or engineering support they need, that makes innovation slower than it should be.”
The ChatGPT for biology
In France, biotech startup Bioptimus is building a foundation model described as the ChatGPT for biology. It’s capable of understanding and engineering multi-modal biology at scale, from DNA sequences and protein structures, to cellular interactions and phenotypes.
We need to build better bridges between the scientific, technological and clinical communities
Historically, this research has focused on isolated biological components, but Bioptimus brings every layer together to simulate how biology exists in reality.
Since launching in 2024, the founding team has raised $76m in funding. Its first model, H-optimus-0, has had more than 100k downloads from around the world since it was released on Amazon Web Service a year ago.
“There’s a lot of pathology workflow optimisation that we’re seeing happening,” Mathilda Strom, chief operating officer at Bioptimus, says. “People are using it for biomarker discovery, to find new cures for diseases, and making connections we haven’t been able to as humans.”
AI is facilitating a golden age of scientific discovery, Felipe Llinares, cofounder and vice president of AI at Bioptimus, adds.
“The data is starting to be there, the models and training methodology are starting to be there, the hardware has improved a lot,” he says. “But biology is very complicated. Trying to simulate a human being with these models is like trying to predict the weather by modelling how air molecules interact with each other.”
Even so, Bioptimus is making progress. It’s on schedule to release its M-Optimus model later this year.
This will be the first multi-scale, multi-modal level trained not just to interpret medical images, but to understand how they relate to gene expression, spatial location and cellular organisation. It will enable researchers to understand, for example, how the arrangement of cells in tumours may impact a patient’s response to treatment.
It will be a significant step forward for the sector, but one that is only possible with the combination of data and HPC, Llinares says.
“One image of a tumour can be as big as 50k pixels by 50k pixels, so the compute pressure is very high. AI for biology is likely to be at the forefront of Europe’s HPC needs over the next few years.”
That data sovereignty is important in healthcare for compliance reasons, he adds. “The type of data we have is highly sensitive so we’re restricted by which regions in the world this data can live in, and are pressed to find compute capacity here.”
Burkov agrees, adding: “We need to build better bridges between the scientific, technological and clinical communities. Our goal is to empower biotech innovators with the infrastructure they need to go further, faster, so they can solve some of the world’s toughest healthcare challenges.”
Read the orginal article: https://sifted.eu/articles/ai-healthcare-brnd/
