SC
The Quest to ‘Solve All Diseases’ with AI: Isomorphic Labs’ Max Jaderberg
@ Sequoia Capital
04/29/25
Takeaways
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Reinforcement learning allows us to tackle problems where we don't know the answers, focusing on whether the model's outputs are good or not, rather than needing predefined solutions.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The breakthroughs in reinforcement learning at DeepMind, particularly in games like Capture the Flag and AlphaStar, have shown that we can achieve human-level performance in complex tasks.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Demis Hassabis has a unique ability to inspire and energize teams, pushing them to see beyond immediate challenges and envision long-term ambitions in AI development.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Our goal is to solve all diseases by leveraging AI to transform our understanding of biology and chemistry, enabling us to manipulate and craft solutions effectively.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
We aim to create a general drug design engine with AI that can be applied across various disease areas, not just targeting a single indication.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The ambition at Isomorphic Labs is to develop a drug design engine that can continuously adapt and apply to new challenges across different diseases, fundamentally changing drug discovery.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The first generation of AI applications in drug design often focused on local models, but our approach is to build a more comprehensive system that can generalize across various tasks, as demonstrated by AlphaFold 3, which allows us to create models that generalize across chemistry and target space, enabling application to any protein or small molecule without needing fine-tuning.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
AlphaFold 3 represents a significant breakthrough in understanding biomolecular structure, allowing for experimental-level accuracy in predicting protein structures and their interactions with small molecules, DNA, and RNA.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
To achieve transformative drug design, we need several more breakthroughs similar to AlphaFold, focusing on different core concepts of biology and chemistry.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Drug design is complex, requiring an understanding of how small molecules interact with proteins, modulating their functions in the body.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The term 'holy grail model' refers to predictive models that can accurately forecast molecular properties, but even the best models won't solve drug design challenges alone due to the vast molecular space that remains unexplored.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Generative models are essential for navigating the vast molecular space, allowing us to explore potential designs that could lead to effective therapeutics.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
AlphaFold 3 enables drug designers to model not just protein structures but also how these proteins interact with small molecules, enhancing our ability to design effective drugs.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The training of AlphaFold 3 involved overcoming challenges in tokenizing various molecular types, including proteins, DNA, and small molecules, to accurately predict their structures and interactions.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
While there are data constraints in machine learning, we can still make significant progress in biology with the existing data, and there's a massive opportunity to generate new data tailored for machine learning applications.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Synthetic data plays a crucial role in drug discovery, especially in chemistry, where we can use theoretical models to create scalable molecular simulations and enhance the information content of our datasets.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
There's a significant opportunity in generating in vivo data, which is currently limited, to better understand drug interactions in real biological systems, and we're seeing breakthroughs in data-generating technology in biology and chemistry that will significantly impact how we model these fields.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
At Isomorphic Labs, we collaborate with various companies to generate proprietary data, which has proven to be impactful in drug development.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Drug design is complex, with numerous bottlenecks before reaching clinical trials, and we need to innovate in this space to improve the process.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Understanding how molecules interact with all proteins in the body is crucial, as this knowledge can help predict toxicity and improve drug design.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
AlphaFold 3 allows our drug designers to see how their molecule designs interact with protein targets instantly, which is a game changer in drug design.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Before AlphaFold 3, drug designers had to rely on intuition and lengthy lab processes to understand molecular interactions, which is no longer the case.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Our internal drug design projects focus on immunology and oncology, and we've made significant progress using advanced models.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Partnerships with Eli Lilly and Novartis have allowed us to tackle challenging problems in drug design, leading to the discovery of new chemical matter.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Building a diverse team of experts in AI, chemistry, and biology is essential for questioning traditional approaches and fostering innovation in drug design.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The intersection of drug design and machine learning is a new field, and we need to bring together experts from both areas to drive breakthroughs.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
Many team members at Isomorphic Labs come from non-traditional backgrounds, which allows for fresh perspectives and innovative problem-solving.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The launch of the AlphaFold server is crucial for providing the scientific community with access to our models, facilitating non-commercial academic research.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
AlphaFold 3 is a significant step towards solving the structure prediction problem, but we still aim for 100% accuracy in understanding molecular interactions.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
The GPT-3 moment in AI biology will involve generative models recreating biological realities, potentially leading to the discovery of new stable compounds.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
As AI-designed drugs enter clinical phases, we must rethink how we engage with regulatory bodies and streamline the clinical trial process.
MJ
Max Jaderberg
04/29/25
@ Sequoia Capital
In the future, the pharmaceutical industry will integrate AI into drug design, making it an essential tool for biology and chemistry.