Ashty Karim

Accelerated enzyme engineering by machine-learning guided cell-free expression @NatureComms 1. This study presents a novel machine-learning (ML)-guided platform for enzyme engineering, integrating cell-free DNA assembly, expression, and functional assays. It significantly accelerates the process by exploring protein sequence space and optimizing enzymes for diverse chemical reactions. 2. The platform was demonstrated on amide synthetases, evaluating 1217 enzyme variants across 10,953 reactions. Using ML, researchers built predictive models that achieved up to 42-fold activity improvements over parent enzymes for nine pharmaceutical compounds. 3. A key innovation is the use of cell-free gene expression systems in a design-build-test-learn (DBTL) workflow, which bypasses the limitations of traditional directed evolution, such as narrow sequence-space exploration and iterative site saturation mutagenesis. 4. The ML-guided framework includes augmented ridge regression models, which use sequence-function relationship data to predict higher-order mutants. These models showed superior performance, requiring only CPU-level computational resources. 5. This approach enabled rapid creation of specialized biocatalysts, reducing enzyme engineering timelines to just one week for multiple campaigns. For example, a quadruple mutant for moclobemide synthesis achieved a 96% conversion rate with a 42-fold increase in catalytic efficiency. 6. Beyond activity improvements, the engineered enzymes retained strict regio- and stereoselectivity without additional selective pressures, showcasing the robustness of the framework. 7. The study emphasizes the versatility of the platform, applying it to engineer six additional pharmaceutical compounds, yielding activity improvements ranging from 1.6- to 34-fold over the wild type. 8. By combining ML with cell-free systems, the framework circumvents path dependencies and expands the protein engineering toolbox, enabling simultaneous exploration of broad sequence spaces with reduced experimental burden. @MichaelCJewett @akarimlab 💻Code: github.com/grantlandwehr/acc… 📜Paper: nature.com/articles/s41467-0… #MachineLearning #ProteinEngineering #SyntheticBiology #Biocatalysis