AI BioDesign Initiative Aims To Design Proteins Nature Never Produced

Evolution has created an astonishing amount of diversity, shaping much of our understanding of biology. But this diversity, which took billions of years to emerge, barely scratches the surface of what could have existed given the trillions of possible DNA sequences available in nature.

A new research accelerator will combine AI and synthetic biology to chart this unexplored space and build novel biomolecules that don’t currently exist naturally. The goal of the initiative, AI BioDesign, is to model the rules biology uses to build life and enable the engineering of new biological solutions, from drugs for neurodegenerative diseases to plastic-destroying enzymes and more power-efficient biological computers.

The accelerator is a collaboration that brings together the Allen Institute’s experience building large-scale, open-source platforms, the University of Washington’s (UW's) expertise in synthetic biology and genome science, and the Fred Hutch Cancer Center’s knowledge of cellular systems, genomics, and translational medicine. This comes at a time when AI is integrating into many aspects of research, massive datasets are readily available, and cutting-edge genomic technology allows researchers to interrogate more complex forms of human genetic variation.

AI BioDesign, which is supported by the Fund for Science and Technology, will leverage AI to create a platform for biological design built on continuous design-build-measure-learn cycles. AI models will propose new biomolecules, which scientists will then build and test at scale. The results of these experiments will then be fed back into the AI models, creating increasingly more accurate and informative engineering cycles.

This platform has the potential to transform how researchers explore the origin and function of proteins. “For decades, we proceeded from sequence to structure to function. Now we can start with a desired function (binding a target, catalyzing a reaction, etc.,) and work backward, leveraging AI to identify protein structures and ultimately the DNA sequences that are likely to achieve those,” Nobel laureate Prof. David Baker, lead scientific director of AI BioDesign, director of the UW Medicine Institute for Protein Design (IPD), and a Howard Hughes Medical Institute investigator, told Technology Networks.

AI will play a key role in the process, allowing researchers to navigate an unimaginably large design space. “Modern models, like those we develop at the IPD, can learn patterns linking sequence, structure, and function, allowing for the generation of candidate proteins that likely wouldn’t have been found through random screening,” Baker explained.

The potential of “designer” proteins

A very specific set of circumstances led to the evolution of the natural proteins known today. “Modifying them can be a challenge—somewhat like renovating an old building,” stated Baker.

“De novo design is a new build. We can create exactly the structural features needed for a particular task. We can build molecular elements that have no close natural counterpart. Many times, designing from scratch is actually simpler than trying to engineer around evolutionary compromises found in existing proteins,” he added.

Designing novel proteins allows researchers to venture into previously uncharted regions of biological possibility. “By creating new molecular architectures, we can discover what biology is capable of beyond what nature happened to produce under a specific set of evolutionary circumstances,” said Baker.

“One of the most powerful ways to understand biology is to try to engineer it.” — Prof. David Baker.

Baker and his team at IPD have already demonstrated the potential of computationally designed proteins for solving scientific challenges. In a recent study published in the journal Science, they introduced a new way to solubilize membrane proteins without detergents using AI-designed proteins, making some of biology’s most challenging proteins easier to study.

In another study, researchers described the creation of a new class of fluorescent imaging tags called NovoTags. The tags can help locate specific proteins and probe their interactions inside human cells using an array of advanced light microscopy techniques. The project leveraged the AI-driven de novo protein design pioneered by Baker, who was awarded the Nobel Prize in chemistry in 2024 for this innovation.

AI-powered synthetic biology: The next frontier

The applications of AI in research are rapidly expanding. Advances in single-cell sequencing, subcellular imaging, and computational power have led to the development of virtual cells that simulate cellular dynamics and predict biological phenomena. Meanwhile, drug discovery is benefiting from AI’s ability to streamline target identification and molecular design.

AI BioDesign is designed to complement those efforts by creating open and reusable resources that the global scientific community can use to accelerate biological design.

Before any of these AI-designed molecules can be translated into custom solutions for human health, a couple of hurdles must be overcome.

“For a designed molecule to be considered reliable, we need evidence across several levels: that it adopts the intended structure, performs the intended biological function, remains stable under physiological conditions, and continues to work when placed in increasingly complex biological environments,” Baker explained.

The most compelling case that an AI-designed molecule is successful will be its ability to function robustly in cells, tissues, organisms, or real-world applications. A clear understanding of failure and unintended interactions is also vital for painting a broad picture of how a molecule is functioning.

“Our ability to design individual molecules has advanced extraordinarily quickly, but biological systems are more than the sum of their parts. One major challenge is predicting how designed molecules will behave within the complex networks that make up living organisms,” said Baker.

“We also need stronger links between computational design, high-throughput experimentation, and real-world deployment. That will help translate molecular design into medicines, diagnostics, sustainable manufacturing processes, and other new technologies,” he concluded.