Alphabet's Isomorphic Labs: Turning Cancer Into a Chronic, But Livable Disease

About this video

• Video Title: Alphabet's Isomorphic Labs: Turning Cancer Into a Chronic, But Livable Disease
• Channel: Bloomberg Technology
• Speakers: Max Jaderberg, Rebecca Paul, Tom Mackenzie
• Duration: 00:21:42

Overview

This video explores Isomorphic Labs, an Alphabet subsidiary, and its innovative approach to drug discovery using artificial intelligence. The discussion highlights the development of a sophisticated drug design engine, the role of AI models like AlphaFold, the challenges of data integrity and generalizability, and the potential to transform diseases like cancer into manageable chronic conditions.

Key takeaways

• Isomorphic Labs is developing an AI-powered drug design engine capable of generating novel molecule designs for various diseases and modalities.
• The engine relies on a suite of AI models, including those that predict protein structures (like AlphaFold) and binding affinities, as well as generative AI capabilities.
• Traditional drug discovery is a slow, iterative process involving lab work and testing; Isomorphic Labs aims to accelerate this by simulating these steps in a virtual environment.
• Ensuring data integrity and quality is crucial for training AI models, with the company investing in data strategy and generating its own wet lab data.
• Generalizability of AI models is a key focus, enabling their application to novel targets and the discovery of new chemical matter, though this approach is more ambitious and challenging to build.
• The company is focusing on immunology and oncology for its pipeline due to tractable clinical trials and significant disease impact.
• While a complete "cure" for cancer is complex due to its many forms, the goal is to make it a chronic, livable disease with improved patient outlooks.
• In 10 years, medicine may involve AI-driven diagnostics and more personalized, faster treatment development, with Isomorphic Labs collaborating to achieve this vision.

Isomorphic Labs aims to accelerate drug discovery by moving the process from the real world into a virtual, computer-based environment. Instead of the traditional, very stepwise and iterative approach of making a molecule in the lab, testing it, and then iterating, they design and test molecules through AI models on a computer. This allows for multiple iterations virtually before taking the best candidates into the lab, thereby skipping steps and significantly reducing the time frame.

Generalizability is key for Isomorphic Labs because they aim to build a reusable drug design engine, not just one for a single program. They want their AI models to be applicable to any target in any disease area, even against different modalities. This means the models, trained on a certain set of data, can be applied to something completely novel, like a protein or target that has never been worked on before, or to discover new chemical matter. This approach is more ambitious and challenging but holds the potential for more significant outcomes.

Table