NVIDIA · Video summary
NVIDIA GTC 2025: Foundation Models in Biology
Biological foundation models learn from evolutionary data to drive scientific discovery and drug development, moving beyond simple pre-training to complex, multimodal applications.
Watch on YouTube ↗Key takeaways
- 🎯Biological foundation models are best defined as distributional learners that capture the vast data of natural evolution (e.g., protein sequences) to build a generalizable understanding, enabling downstream applications.06:26
- 🚀Focus on identifying and leveraging evolutionary data as a core component when building or evaluating biological foundation models.08:17
- 📈While evolution provides a strong foundation, new data sets beyond natural processes are crucial for training models on aspects like manufacturability or specific drug efficacy, moving towards a more complete picture.08:44
- 💡The power of biological foundation models stems from evolution's unifying theory, which operates across multiple scales (ecosystems to molecules) and provides a rich, inherent structure for AI to learn from.09:50
- 🚀The ultimate utility of these models lies in their application, enabling mechanistic discovery, drug development, and creating an 'app store' for biology, rather than just being general pre-trained models.10:51
- ⚡Current biological models often feel like communicating with a heavy accent (e.g., DNA sequences), necessitating new methods to bridge the gap between model-generated language and human understanding for effective iteration and reasoning.11:23
Chapters
- 0:00🎤 Introduction and Housekeeping
- 1:48👨🔬 Panelist Introductions and Backgrounds
- 4:31🔍 What Are Biological Foundation Models?
- 7:35💡 Comparison with Other Fields
- 8:11🧬 Evolution and Biological Data
- 9:45🌐 Universal and Generalist Models
- 10:48🧪 Applications and Challenges
- 14:33📝 Evaluating and Building Data for Models
- 16:19🔄 Data Generation and Model Improvement
- 16:43🌉 Multimodal Foundation Models
The digest
High-Level Overview
This session presents a comprehensive exploration of foundation models in biology, discussing their definition, applications, key challenges, and future directions. The discussion clarifies what it means to build generalizable AI models for complex biological data, contrasting them with foundation models familiar from natural language processing (NLP) and vision. The panel features leading experts from biotech, academia, and AI research who explore how these models learn from vast biological datasets, leverage evolutionary principles, and aim to impact fields such as drug discovery, synthetic biology, and scientific research. This briefing is essential for professionals involved in computational biology, bioengineering, AI-driven drug discovery, and life sciences innovation seeking a deeper conceptual and practical understanding of foundational AI in biology.
Keep reading in The Digest
The full 5-minute article, 9 more sections, and every new episode from NVIDIA and the creators you follow.
- Structured Deep Dive
- 1. Introduction to Biological Foundation Models (00:30 – 11:00)
- 2. The Role of Evolution as a Foundational Principle (07:00 – 10:00)
- 3. Comparison with Chemistry and Multimodal Challenges (11:00 – 15:00)
- 4. Data: Quantity, Quality, and Iterative Learning (14:00 – 16:40)
- 5. Practical Applications and Future Directions
- Key Framework: Defining a Biological Foundation Model
- Practical Implications & Takeaways
- Action-Oriented Summary
More from NVIDIA
Advancing Infrastructure for the Era of Agentic AI | Ian Buck at AI Infra Summit 2026
NVIDIA's Ian Buck discusses the evolution of AI infrastructure, highlighting the shift towards agentic AI and the full-stack Vera Rubin platform designed to meet its demanding computational needs.
The Infrastructure of Intelligence: NVIDIA and OpenAI on a Decade of Building Frontier AI
The AI Infra Summit brought together industry leaders to discuss the challenges and innovations in building AI infrastructure at scale. The conversation delved into the evolving definition of scale, the complexities of designing compute token factories, and the pivotal role of partnerships in navigating this unprecedented growth.
How Mistral Is Building Frontier AI for the Enterprise | NVIDIA AI Podcast Ep. 301
In this episode of the NVIDIA AI Podcast, Tim Lacroix, co-founder and CTO of Mistral AI, discusses the company's commitment to open source models, their collaboration with NVIDIA, and the development of their new platform, Mistral Forge, aimed at enhancing enterprise AI solutions. Lacroix emphasizes the importance of customization in AI to meet specific industry needs while fostering global innovation.
NVIDIA GTC 2026: Healthcare Special Address
In this NVIDIA GTC 2026 address, NVIDIA discusses the transformative impact of AI in healthcare, highlighting advancements in digital labs, agentic AI, and robotics that enhance disease understanding, drug discovery, and patient care. The presentation emphasizes collaboration and the integration of AI technologies to create a more efficient and accessible healthcare system.
Automotive Special Address: Advancing Level 4 Autonomy, the Path to Scalable, Safe AVs and Robotaxis
NVIDIA's Drive AV platform is evolving with physical AI and a three-computer architecture (training, simulation, inference) to accelerate Level 4 autonomy through hardware, software, and open-sourced models.
NVIDIA GTC 2026 Open Models Panel Highlights with Jensen Huang
AI is evolving beyond single models to complex systems and agents, with open models playing a crucial role in driving innovation, customization, and trust, especially for mission-critical applications.