The dirty secret of AI-powered protein design is that the models are getting very good at predicting what enzymes look like, and rather less good at predicting whether they will actually do anything useful. Imperagen, a spin-out from the University of Cambridge, believes it knows why: the field has been ignoring quantum mechanics.
The company announced this week that it has raised £5 million to build what it calls a "quantum-accurate" enzyme engineering platform. The thesis is straightforward, if technically ambitious. Enzymes catalyse reactions through quantum tunnelling and other subatomic phenomena that classical molecular dynamics cannot capture. Current AI tools—including the celebrated AlphaFold family—excel at structure prediction but treat the quantum layer as a rounding error. Imperagen argues this is why so many computationally designed enzymes fail when they hit a wet lab.
The gap between prediction and function
AlphaFold and its successors have transformed structural biology, but structure is not destiny. An enzyme's catalytic power depends on the precise choreography of electrons at the active site, a domain where quantum effects dominate. Imperagen's platform combines quantum chemistry simulations with machine learning to model these effects at scale, aiming to predict not just what an enzyme will look like but how fast it will turn substrate into product.
The commercial implications are significant. Industrial enzymes already represent a market worth tens of billions of dollars annually, underpinning everything from laundry detergent to pharmaceutical synthesis. If Imperagen can reliably design enzymes with specified catalytic rates—rather than merely plausible shapes—it could compress development timelines that currently stretch for years.
Why now
Two converging trends make this moment plausible. First, quantum chemistry software has matured enough to run on classical hardware at meaningful scales, sidestepping the need for actual quantum computers. Second, the explosion of protein language models has created a foundation of learned structural knowledge that can be augmented with physics-based constraints. Imperagen is essentially proposing a hybrid: let AI handle the coarse-grained search, then let quantum simulations adjudicate which candidates will perform.
The £5 million seed round was led by Cambridge Enterprise and joined by undisclosed life-science investors. The sum is modest by AI standards but substantial for a company whose core asset is a scientific hypothesis rather than a product. The funds will support hiring computational chemists and expanding partnerships with industrial enzyme users.
Our take
Imperagen is making a contrarian bet that the AI-protein revolution has been too superficial—that the field's obsession with structure has obscured the quantum reality of catalysis. If the company is right, it could become the physics layer beneath a generation of biotech startups. If wrong, it will join a long list of academic spin-outs that overestimated how quickly deep science translates into deep markets. Either way, the wager is worth watching: the gap between predicting proteins and engineering them remains one of the most consequential unsolved problems in applied AI.
