Google Deepmind just broke chemistry…
They open-sourced a model that can invent completely new biology from scratch.
It’s called “AlphaProtein Novo”
For decades, scientists have dreamed of designing enzymes—nature’s tiny molecular machines—from scratch to perform chemistry that doesn’t exist in the natural world. Traditionally, protein engineers have had to start with enzymes that already exist in nature and tweak them, but finding the right natural starting point is often difficult, and there’s no guarantee nature has an enzyme for the job you want done.
A new approach called de novo design aims to build enzymes from first principles, like designing a tool for a specific task rather than adapting whatever happens to be in the toolbox. Until now, however, designed enzymes haven’t been good enough to be practically useful.
Enter AlphaProtein Novo (AP Novo), a machine-learning pipeline that changes the game. The researchers show, for the first time, that designing enzymes from scratch can actually beat searching through nature’s existing enzyme diversity when it comes to tackling hard chemistry problems.
Using AP Novo, the team designed entirely new enzymes for two impressive tasks. First, they created “nitrene transferases”—enzymes that don’t exist in nature—to build piperidine, a molecular building block important in many medicines, with exceptional precision in producing the desired form of the molecule. Second, they designed enzymes that can break down DEHP, a harmful environmental pollutant, under harsh conditions that would destroy natural enzymes.
They also achieved top-tier performance on two well-studied benchmark reactions. By analyzing what made these designs work, they discovered that two strategies were key: using predictions from AlphaFold 3 (an AI system that predicts protein structures) to guide designs based on chemical mechanisms, and scoring candidate enzyme scaffolds by looking at how ensembles of related sequences behave.
Overall, the study shows that de novo design has matured into a powerful tool that complements—rather than replaces—nature’s own enzyme repertoire, opening the door to custom-designed catalysts for all sorts of applications.
#ArtificialIntelligence #denovodesign #proteinengineering #machinelearning #DrugDiscovery
Creating highly active enzymes for arbitrary reactions is a transformative goal for the molecular sciences. Traditional protein engineering (2,3) is constrained by a reliance on existing natural starting points, which are often challenging to discover. De novo design offers the potential to overcome this by creating enzymes from first principles , but has yet to achieve practically relevant catalytic properties. Here we present AlphaProtein Novo (AP Novo), a machine-learning pipeline that demonstrates, for the first time, that de novo enzyme design can outperform natural sequence mining in addressing challenging chemistry. We used AP Novo to design new-to-nature nitrene transferases to synthesise the pharmacophore piperidine with unprecedented product selectivity, and to create enzymes that degrade the environmental toxin DEHP under conditions that denature natural enzymes. We also obtained state-of-the-art catalytic efficiencies on two well-studied model reactions, and through iterative design and analysis found that key enablers of success were mechanism-inspired metrics based on AlphaFold 3 predictions and sequence-ensemble-based scoring of designed backbones. Our results show that de novo design has become a powerful complement to natural enzyme diversity for discovering catalysts for a range of applications.
Author-affiliated entities have filed a US provisional patent application relating to the de novo design of proteins and enzymes using generative diffusion models and structure prediction neural networks. All of the authors other than Z.-Q.L., M.D., A.N.M., J.C.R., Y.Z., P.L. and F.H.A. have commercial interests in the work described.