In the rapidly evolving landscape of AI for science, Chai Discovery is making significant waves. This protein design startup, just two and a half years old, is fundamentally changing how new medicines are conceived and developed. Co-founders Matt McPartland and Neil Patil recently joined the Latent Space podcast to discuss their ambitious mission, detailing how their AI models are transforming biology into a more agile, software-like process. This approach allows for chai discovery designing proteins like software, moving beyond traditional biological research methods.
From Computer Science to Biology: The Founders' Journey
Matt McPartland, co-founder at Chai Discovery, brings a background rooted in theoretical computer science, which he transitioned to AI biology during his PhD. He witnessed firsthand the transformative power of AI in protein structure prediction, starting from the early days of AlphaFold. "I was in the field during AlphaFold 2 and like got to see a lot of the interesting developments at that time," McPartland shared. He saw the potential to apply these advancements in the real world, a vision that led to the founding of Chai.
Neil Patil, who leads platform and product at Chai, has a more varied background, spanning app development, robotics, and SaaS before finding his way to Chai. "I kind of got into programming like 15 years ago, making apps in the app store. Got really addicted to the dopamine hits you get from that," Patil explained. He later transitioned to robotics and self-driving cars, eventually finding his way to Chai about a year ago to help scale its platform and commercialization efforts.
Chai's Core Thesis: A Software Factory for Medicines
Chai's core thesis is to act as the software and modeling layer for drug discovery, a strategy that was once controversial but is now proving its value. "Drug discovery is a very lengthy process, right? And a lot of these pharma companies are spending lots of time, you know, years and years and billions of dollars trying to find initial therapeutic candidates," Patil noted. Chai's AI models are designed to accelerate this by identifying initial binders and beyond.
Unlike companies focused on developing their own drugs, Chai positions itself as a "neutral software factory for making medicines." This approach allows them to partner with and support various pharmaceutical companies in their drug discovery journeys. The company has already announced significant partnerships with major players like Eli Lilly, Pfizer, Novartis, and Genentech, demonstrating the compelling nature of their offering to investors and customers alike. This focus on a software-centric approach to drug discovery is a key aspect of chai discovery scaling drug design software.
A "Photoshop for Molecules" Interface
McPartland described Chai's platform as looking "a lot less like a chat GPT and a lot more like uh Autodesk or or Solid Works or or Figma." Users can load up their molecules and utilize tools analogous to Photoshop's, such as a "paint tool to kind of paint your epitope" and a "content-aware fill tool to kind of get your uh your binders generated from Chai." This intuitive, visual interface is crucial for researchers navigating the complex world of protein design.
From Waterfall to Agile: Revolutionizing Drug Discovery
Traditionally, drug discovery has followed a "waterfall model" with distinct phases like target discovery, hit discovery, and optimization, each taking months to years. This process is costly and slow, making early-stage experimentation expensive. However, Patil highlighted how AI is changing this: "If you start to get in a regime where you can have models give you really promising candidates, you can start to make that look a lot more like a loop." He likened this shift to becoming "more agile in software development," enabling faster iteration and discovery.
Precision Design: Focusing on Antibodies and Beyond
Chai's focus on antibodies is strategic. Antibodies, with their Y-shaped structure and specific binding tips (CDR loops), offer flexibility in design. While predicting their binding is complex, the ability to design them allows for greater control. "You can be a lot more selective about the types of designs you want to make and the types of structures you actually want to focus on," McPartland explained. This precision allows for targeted therapies, such as Antibody-Drug Conjugates (ADCs), where a toxic molecule is delivered directly to cancer cells.
The company is also pushing the boundaries into more complex therapeutic modalities like bispecifics and ADCs, and exploring agonist behavior, which involves precisely "pressing a switch" on cellular targets. "I think like one of the things that's really exciting about where we're getting to with some of these models is we can start to get that precise, right? We can really target a very specific epitope," McPartland added.
The Evolution of Chai's AI Models
The journey began with Chai 1, a structure prediction model that essentially mapped amino acid sequences to their folded 3D shapes. This was followed by Chai 2, an all-atom diffusion model capable of designing new molecules that bind to target structures. This marked a significant leap, enabling generative design for therapeutics. The company is now working on Chai 3, continually improving model accuracy and expanding capabilities.
The progress in protein structure prediction, from AlphaFold 2 to multi-chain predictions and inverse folding, has been rapid. "We didn't have a multi-chain structure prediction model until like 2021. That was 5 years ago when we could like start with deep learning to like actually predict the shape of two proteins at once," McPartland recalled.
Ultimately, Chai's vision is to harness AI's power to accelerate scientific discovery. By building robust models and user-friendly platforms, they aim to unlock new frontiers in medicine and beyond, making complex biological challenges more tractable and driving the future of AI for science, a mission championed by innovators like those at StartupHub.ai.
tags: ai, drug discovery, protein design, biotechnology, machine learning, healthcare, startups
Top comments (0)