- Based in San Francisco and London, Prima Mente is building AI-native infrastructure for molecular biology and neurodegenerative disease research. Its models integrate genomic, epigenomic, and proteomic data to power biomarker discovery, diagnostics, and translational R&D
- The company announces its first family of models, Pleiades. Focussing on human epigenetics, the model accurately determines cell type of origin, uncovers disease biology from cfDNA, and demonstrates scalable in silico biomarker discovery to unlock new capabilities in AI-driven molecular medicine
Prima Mente today announced the launch of Pleiades, a family of large-scale foundation models created to understand the epigenomic and molecular landscape of human disease. Pleiades represents generative transformers (90M, 600M, and 7B parameters) trained on 1.9 trillion tokens at single-nucleotide resolution. The work is introduced in a newly released preprint, authored in collaboration with researchers from University of Oxford, Hebrew University of Jerusalem, and University of Gothenburg.
Trained on a methylation atlas of human cell types as well as cell-free DNA (cfDNA), Pleiades represents a new era in Alzheimer’s research and translational medicine – offering the ability to detect disease-relevant signals from blood with unprecedented resolution across cellular subtypes, identify molecular profiles for experimental design, and reveal targets for early intervention.
“We believe epigenetics and cfDNA offer one of the most powerful, underutilized entry points into understanding the brain,” said Ravi Solanki, CEO of Prima Mente. “They give us a non-invasive lens into cellular activity and dysfunction – particularly in neurodegeneration – at a resolution that was unimaginable just a few years ago. With Pleiades, we’re building the AI infrastructure to translate that signal into meaningful discovery, diagnosis, and eventually, intervention.”
In contrast to traditional models, Pleiades uses a large-scale transformer-based architecture to learn directly from raw sequencing data. It predicts the cell type of origin of cfDNA fragments, uncovers tissue and disease-specific methylation patterns, and generates synthetic cfDNA to power in silico experimentation. Critically, the team demonstrated that model performance scales reliably with data and compute, extending AI’s scaling laws seen in large language and vision models to biology and laying the foundation for future multiomic modeling.
The model has so far been applied to a variety of biological tasks, including tissue classification, disease detection, and fragment reconstruction, allowing it to generalize across tissues and diseases while learning shared representations of molecular function. In early Alzheimer’s work, Pleiades has shown the ability to detect neuron- and microglia-, and T cell-derived cfDNA signatures that signal neurodegeneration before clinical symptoms appear.
“Pleiades represents an important step forward in how we study Alzheimer’s disease using blood-based biomarkers,” said Professor Henrik Zetterberg, a leading researcher in neurodegeneration at the University of Gothenburg. “Understanding the cell type origin and regulatory disruption at this level could enable earlier, more accurate diagnosis and reveal new biological pathways for intervention. I was thrilled to see the performance of the approach alongside the protein-based biomarkers—so much exciting research can be done and the technology really opens for personalized medicine approaches finding neurodegenerative disease subtypes with varying disease drivers.”
Prima Mente has partnered with NVIDIA to scale training across DGX infrastructure, ensuring high-performance compute can be directed toward decoding the biology of complex disease. Additionally, to train and test Pleiades, Prima Mente worked with Nebius, Siam AI, and Eternis Labs for distributed compute infrastructure. The teams will continue to collaborate together on data collection and future training initiatives.
Pleiades caps a period of strong momentum for Prima Mente. Alongside the model’s release, the company built a wet lab for high-throughput multiomic data generation and has begun modeling disease progression across regulatory, transcriptomic, and proteomic layers. The company is launching a 1,000-patient Alzheimer’s study next month and is actively building partnerships across AI, academia, and life sciences to translate these models into tools for discovery, diagnosis, and ultimately, intervention.
Read the full Pleiades preprint
Learn about Prima Mente’s research
About Prima Mente
Prima Mente is based in San Francisco and London. Its mission is to build intelligent infrastructure to deeply understand the brain, protect it from neurological disease, and enhance it in health.
View source version on businesswire.com: https://www.businesswire.com/news/home/20250716401893/en/