How AI is mapping the tree of life | Microsoft Azure and NVIDIA

Key Concepts

• Biodiversity Digitization: Converting biological material into data for analysis and application.
• Foundation Models in Biology: AI models capable of understanding and predicting the function of any protein, gene, or genome in any context.
• Scaling Laws in Biology: The principle that increasing data, compute, and model size leads to improved performance in biological AI models.
• Data Wall: The limitation in model performance due to a lack of diverse and representative data.
• Evolutionary Modeling: Using AI to model and guide protein evolution for desired features.
• Bioeconomy: Economic activity derived from biological and renewable resources.

1. The Challenge of Digitizing Biodiversity

• The core mission is to digitize nature to understand evolved solutions for human benefit, bridging biotechnology and biodiversity.
• A spoonful of soil can contain as many living organisms as there are humans on Earth, representing a vast, largely unknown biological space.
• The current knowledge of biology is estimated to be 17 to 20 orders of magnitude less than what remains unknown, likened to five drops of water compared to the Atlantic Ocean.
• "Wildly, we've only just scratched the surface of what humans know about biodiversity."
• The goal is to build a data asset that representatively depicts the diversity that exists out in nature.

2. Data Acquisition and Processing

• Data collection involves going to where biodiversity exists, such as Costa Rica, which contains approximately 6% of the world's biodiversity within 0.03% of the Earth's surface.
• The process involves converting biological material into data, specifically DNA sequencing.
• Computational workflows begin after DNA extraction and sequencing, involving complex processes to make natural world data machine-readable.
• Data processing includes bioinformatics assemblies, annotations, and labeling with biological and bioinformatics tools.
• "Essentially what we're doing is we're converting biological material to data."

3. Overcoming the Data Wall with Scaled Data

• Current protein language models are plateauing due to a lack of diverse data from public databases.
• The solution is to build a comprehensive data asset that accurately represents the diversity found in nature.
• Scaling laws are expected to hold true in biology, allowing for transformational improvements in biotechnology and healthcare with increased data.
• The aim is to grow the database by 10x each year, presenting a significant engineering challenge.
• "I think one of the things I really started to appreciate is that biology is hitting a data wall."

4. AI and Foundation Models in Biology

• AI models are used to interrogate data and build representations of it.
• The goal is to develop foundation models that can understand any protein, gene, or genome in any context, a capability currently lacking in biology.
• These models require foundational datasets that cover every corner of the design space.
• AI models are used to evolve proteins to have desired features, reducing the need for millions of variants.
• "We want to see foundation models explode in biology that can understand not just one protein's function, but potentially any protein or any gene, any genome in any context."

5. The Role of Compute and Partnerships

• Data processing requires hundreds of millions of CPU hours, with billions of annotations at the protein and DNA levels.
• Partnerships with Microsoft (Azure) and NVIDIA are crucial for scaling models and compute.
• Azure Kubernetes Service (AKS) and container registry are used to orchestrate and scale annotation pipelines.
• NVIDIA GPUs and the BioNeMo team are used to optimize architectures for the company's unique data.
• "At the scale of the data that we're looking at, that needs to be done by hundreds of millions of CPU hours."

6. Understanding the Language of DNA

• DNA contains a complex language that determines an organism's functions.
• Understanding this language is crucial for reprogramming novel proteins.
• This requires data and compute to understand how proteins recognize their sites.
• "There's this inherent language around what a strand of DNA will let an organism do."

7. The Importance of Context

• Context matters in biology for all applications and tools.
• A comprehensive view of sequences, genomes, and their evolution is necessary for understanding biological systems.
• Environmental factors, such as changes in moisture or competition, can drive rapid genomic changes in bacteria.
• "Context in biology matters with every single application and every single tool that we want to use."

8. Applications and Impact

• The work has applications in finding gene editing proteins for human therapeutics and other purposes.
• AI tools can help discover new functions and design novel proteins.
• The company aims to provide data and insights to pharma, agriculture, and chemical companies.
• The ultimate goal is to enable the control and design of biology for various applications.
• "All of these amazing things that we want in the world come down to our ability to control biology and design biology, and computers are how we do that."

9. Collaboration and Global Impact

• The company collaborates with a global network of partners, including researchers and institutions.
• They enhance capacity in bioeconomy regions by building labs, sharing data, and training scientists.
• Revenue is shared with partner countries to demonstrate the value of biodiversity data.
• The project with CENIBiot in Costa Rica exemplifies the potential for collaboration between public and private institutions.
• "Biodiversity is not just about conservation. It is a source of innovation."

10. Conclusion

Basecamp Research is pioneering work in digitizing and understanding biodiversity, leveraging AI, and fostering global collaboration. The company's mission is to overcome the data wall in biology, enabling the development of new biological products and solutions across various industries. The focus on scaling data, compute, and models, combined with a commitment to ethical and collaborative practices, positions the company at the forefront of the bio-revolution.