Objective
Prima Mente, an AI biology company, builds and applies world-leading biological foundation models. Their first goal is to understand the brain across health, age and disease to cure severe human diseases like Alzheimer’s. The London and San Francisco-based startup recently built the world’s first whole-genome epigenetic foundation model—Pleiades—driving breakthroughs in precision diagnostics and AI-first neuroscience research using the NVIDIA DGX™ Cloud Lepton AI platform running on the Nebius AI Cloud.
Customer
Prima Mente
Partner
Nebius
Use Case
NVIDIA Data Center / Cloud
Key Takeaways
Enhanced DNA Generation
- The AI foundation model, Pleiades, generated DNA fragments with 83% accuracy using blood samples.
Faster, Accurate Disease Detection
- For early detection of Alzheimer’s and Parkinson’s, Pleiades achieved 0.89 accuracy, rising to 0.97 with protein tests.
cfDNA Enrichment Capability
- Pleiades can increase the amount of DNA from specific cell types in a blood sample by up to 20x, making it easier to detect important signals with minimal input.
Bridging the Epigenetic Gap in Brain Science
Aging and neurological diseases like Alzheimer’s and Parkinson’s can’t be fully understood by looking at DNA alone. These diseases develop gradually, with early indicators including minor memory lapses for Alzheimer’s and subtle movement changes for Parkinson’s, making early detection especially challenging.
Epigenetics refers to how environmental factors and behaviors can influence gene expression without changing the underlying DNA sequence. Epigenetics helps determine which genes are turned on or off, showing how these shifts can impact health, including how neurological diseases like Alzheimer’s and Parkinson’s develop or progress. That’s because important signals in the brain come from epigenetic modifications such as DNA methylation—a chemical change that regulates gene activity. Yet these signals are often difficult to detect using DNA-only approaches, and require the unique chemical modifications that enable study of the epigenome like methylation.
Prima Mente, an NVIDIA Inception member, set out to bypass these limitations and build powerful AI foundation models that could learn from both the whole genome and epigenome, across a multitude of human cell types. Launched in July 2025, the Pleiades family of foundation models introduces a new paradigm in precision diagnostics by jointly analyzing DNA and epigenetic patterns.
Pleiades combines high-resolution genomic data with advanced AI architecture, outperforming traditional genome models and opening new possibilities for the early detection and treatment of complex brain disorders.
Infrastructure Built for the Future of Neuroscience
Modeling the whole human genome, particularly the chemical language of DNA methylation, demanded an innovative approach. Working in collaboration with global research institutions like the Department of Clinical Neurosciences, Experimental Psychology, and Neurology at the University of Oxford, Prima Mente curated vast human methylation and genomic datasets to pretrain and fine-tune Pleiades.
Available in 90M, 600M, and 7B parameter model variants, Pleiades is built on an autoregressive transformer decoder architecture. It’s a specific type of neural network architecture, where a model generates one token at a time and uses its own previously generated output as input for the next prediction. Pleiades has been trained using 1.9 trillion tokens of epigenetic data. The Pleiades models are the first foundation models trained on methylation data at this scale.
Each model was rigorously evaluated against real-world clinical test sets and optimized bio-benchmarks, demonstrating consistent performance gains over DNA-only systems across a range of molecular and diagnostic tasks.
Nebius data center in Finland, powered by NVIDIA GPUs
Breakthrough Accuracy in Brain Diagnostics
Even the smallest Pleiades variant outperformed leading DNA-only models across a range of biological and clinical benchmarks.
In generative tasks, Pleiades can expand the usable information in cell-free DNA (cfDNA), the small fragments of DNA that circulate in the bloodstream, by 10x. It reconstructs these fragments with an average per-nucleotide accuracy of 83% and shows a 91% match to the original CpG methylation patterns, the key regulatory markers that control gene activity. Pleiades can also increase the concentration of specific cell types in these fragments by up to 20x, enabling reliable detection of rare cellular signals even when starting from very limited material.
And in disease prediction, the 7B parameter model scored 0.89 for early-stage detection of Alzheimer’s and Parkinson’s using blood samples, rising to 0.97 when combined with protein tests. With 1.0 representing perfect accuracy, these results show Pleiades’ potential for precise, clinical diagnostics and prognostics.
Pleiades opens new frontiers in mechanistic biology, regulatory genomics, and cfDNA analysis—well beyond traditional classification tasks. These breakthroughs represent only the beginning of an AI-first future for medicine, with the potential to support new diagnostic, prognostic, and therapeutic discoveries.
To push forward this work, Prima Mente is actively piloting the Pleiades model in a live clinical study called SANDBOX across 15 sites in the United Kingdom. Assessing patients with suspected neurodegenerative conditions, Pleiades will help classify and stratify patients in order to improve clinical outcomes and manage waiting lists in the UK National Health Service. “Diagnostic and prognostic care will rapidly improve with the use of AI in clinical practice, and, with Prima Mente, I hope SANDBOX is just the start,” said Dr. Ivan Koychev, Consultant Neuropsychiatrist at Imperial College Hospital NHS Trust and Principal Investigator of the study.
“We are embarking upon one of the world’s hardest challenges. Nebius, an NVIDIA Cloud Partner, provided us with deep technical support and exceptional reliability so that we could focus on improving architectures, training, and experimental work. They truly became a partner we can trust.”
Ravi Solanki
Cofounder, Prima Mente
Accelerating Genomic Research With AI
To deliver an AI infrastructure that could handle its massive datasets, support rapid experimentation cycles, and scale seamlessly, Prima Mente partnered with NVIDIA Cloud Partner Nebius to leverage the DGX Cloud Lepton AI platform. This collaboration provided Prima Mente with solutions to transform genomic research through accelerated AI development, robust data sovereignty, and stringent compliance with regional data governance standards.
Using Nebius’ distributed shared-file system and managed Kubernetes clusters, Prima Mente launched multi-host training on highly available NVIDIA supercomputing clusters. The Pleiades 7B model was trained on 256 NVIDIA H200 GPUs and interconnected via the NVIDIA Quantum-2 InfiniBand platform. Nebius’s architects helped set up Prima Mente’s workloads, providing expertise, including guidance on debugging how its pipelines use node interconnect via NVIDIA NCCL.
Working closely with NVIDIA, Prima Mente used DGX Cloud Lepton to fine-tune and validate the 7B-parameter Pleiades model. The team leveraged Lepton’s dev pods and batch job orchestration to manage multi-terabyte genomic datasets, run large-scale experiments, and track results with built-in observability tools. This developer-first workflow enabled rapid iteration, seamless integration with custom Docker environments, and consistent model performance across distributed GPU infrastructure.
NVIDIA H200 Tensor Core GPUs.
By eliminating bottlenecks in high-speed dataset streaming and file storage, Prima Mente was able to:
- Run dozens of experiments simultaneously with shared data and checkpoints
- Dynamically allocate computing resources to each training phase
- Effortlessly reuse model checkpoints for rapid iteration
NVIDIA and Nebius delivered the expertise Prima Mente needed to accelerate its research, from model pretraining to large-scale inference on real-world patient samples.
The Pleiades series is the first set of models to capture methylation changes to DNA due to aging and disease, setting the stage for innovation in early disease detection and therapeutic discovery. With the help of Pleiades and future epigenomic models, neurodegenerative diseases could soon be diagnosed and monitored using a simple blood test.
For now, Prima Mente is advancing this work by developing multimodal models to uncover new biology and potential therapeutic targets for these diseases.
Explore the Pleiades foundation models and how NVIDIA accelerated computing on Nebius is powering brain science breakthroughs.