The ultimate dream of synthetic biology is to make the engineering of biology as predictable and standardized as the engineering of electronics or machines. For proteins, this means moving toward a future where we can design and build them from a set of well-characterized, interchangeable parts, much like snapping together LEGO bricks. For years, this has been a distant vision, as the rules for connecting protein domains are incredibly complex.

The latest generation of compositional and generative AI is finally bringing this modular dream within reach. Models like ProtComposer are pioneering this approach by allowing designers to sketch out a protein’s architecture using a simple layout of 3D shapes, or “ellipsoids”. Each shape can be thought of as a specialized LEGO brick. The designer arranges the bricks, and the AI acts as the master builder, figuring out the complex atomic details needed to connect them into a stable, functional whole. Similarly, flow-based models like Proteina offer hierarchical control, letting designers specify the type of fold or secondary structure, akin to choosing a specific category of LEGO piece to work with.

This is a profound shift in how we approach protein engineering. It opens the door to creating vast libraries of modular protein parts that can be combined in near-infinite ways to create custom enzymes, biosensors, or therapeutics. Imagine ordering a protein from a catalogue of functional domains, specifying their spatial arrangement, and having an AI deliver the final sequence. We are at the very beginning of this journey, but the tools are now in place. What custom proteins would you build for medicine, industry, or the environment if you had a molecular LEGO set at your fingertips?