We are delighted to announce that Bilobe has joined the Nvidia BioNemo ecosystem, gaining access to Nvidia's specialised drug discovery and molecular biology AI platform and to the HiPerGator supercomputer at the University of Florida — one of the most powerful academic computing systems in the United States.
The Announcement
This collaboration represents a significant step forward for Bilobe's research capabilities in molecular medicine AI. The partnership was formalised in November 2024 and builds on our existing research collaboration with the University of Florida, through which we are co-developing the ViraLingo pan-viral language model for variant prediction.
Access to Nvidia BioNemo and HiPerGator enables us to train larger models on larger datasets, run experiments that were previously computationally infeasible, and accelerate the development timeline for several of our most ambitious research projects.
What is Nvidia BioNemo?
BioNemo is Nvidia's platform for AI-accelerated drug discovery and molecular biology. It provides a collection of pre-trained biological foundation models — including models for protein structure prediction, protein language modelling, DNA sequence analysis, and molecular property prediction — along with the GPU-optimised training and inference infrastructure needed to work at scale.
For Bilobe, the most immediately relevant BioNemo capabilities are the protein language models (based on ESM-2 architecture) and the DNA/RNA sequence models, which we are integrating with our ViraLingo pipeline to improve viral protein fitness prediction. BioNemo's pre-trained embeddings provide a strong initialisation point that significantly reduces the amount of viral sequence data needed for fine-tuning.
HiPerGator Supercomputer Access
HiPerGator is the University of Florida's flagship high-performance computing system, currently featuring over 120,000 CPU cores, more than 1,400 Nvidia A100 GPU cards, and 2.5 petabytes of high-speed storage. For AI research, the GPU cluster is the critical resource — A100 GPUs are the gold standard for training large transformer models, and having cluster-scale access means we can run distributed training runs that would be prohibitively expensive on commercial cloud compute.
Our initial allocation under the partnership covers 500,000 GPU-hours over eighteen months — sufficient to complete the ViraLingo pre-training runs and conduct the systematic ablation studies needed to understand which aspects of the pan-viral training objective contribute most to model performance.
"Access to HiPerGator removes the compute constraint that was the main bottleneck on our most ambitious research. We can now ask questions that were previously out of reach."
— Enea Parimbelli, CEO Bilobe
Research Roadmap
The BioNemo partnership enables three research directions that were previously constrained by compute availability:
- ViraLingo scale-up: Training larger versions of the ViraLingo model (moving from 125M to 650M parameters) to test whether model scale improves variant fitness prediction.
- Multi-modal molecular models: Combining sequence-based and structure-based representations for viral proteins, using BioNemo's protein structure prediction capabilities as a source of structural features.
- Drug-target interaction modelling: Exploring whether ViraLingo's viral protein representations can be used to predict binding affinity for antiviral drug candidates — a potential route from variant prediction to therapeutic design.
Key Takeaways
- Bilobe has joined the Nvidia BioNemo ecosystem for molecular medicine AI research
- Partnership includes access to BioNemo foundation models and HiPerGator supercomputer (1,400+ A100 GPUs)
- Initial allocation: 500,000 GPU-hours over 18 months
- Primary application: scaling up ViraLingo pan-viral language model training
- Longer-term directions include multi-modal molecular models and drug-target interaction prediction
Looking Ahead
We are deeply grateful to Nvidia and to the University of Florida's Research Computing team for making this collaboration possible. This partnership reflects Bilobe's dual identity as a healthcare AI company: deeply applied in our clinical work, and deeply committed to the frontier research that makes the applied work possible.
We will publish updates on our BioNemo research through this Insights channel and in peer-reviewed venues as results mature. If you are a researcher or clinician interested in collaborating on molecular medicine AI, we would love to hear from you at info@bilobe.ai.
