Machine learning (ML) is dramatically changing the way enzymes are engineered, making the process faster and more efficient. A new study has unveiled a machine learning-guided framework integrated with cell-free expression systems to expedite the exploration of fitness landscapes across various enzyme sequences. This innovative method holds the potential to optimize enzymes for distinct chemical reactions, broadening the scope of applications from pharmaceuticals to materials science.
Conducted by researchers studying the amide synthetase enzyme McbA derived from Marinactinospora thermotolerans, the study demonstrates how this new ML-guided system can evaluate substrate preferences for over 1,200 enzyme variants across more than 10,900 unique reactions. The findings indicate significant improvements in enzyme activity, with certain ML-predicted variants exhibiting up to 42-fold increases compared to the original enzyme.
The process of enzyme engineering usually involves tedious and slow methods, often requiring directed evolution, which maps the relationship between enzyme sequences and their functions within narrow regions of chemical space. Current approaches can yield poor predictive capabilities due to limited datasets and the singular focus of selection methods which prioritize the best-performing variants for only one transformation.
The study introduces an efficient design-build-test-learn (DBTL) workflow, which incorporates cell-free gene expression (CFE) and machine learning models to navigate fitness landscapes much more rapidly. This approach allows for simultaneous evaluation of multiple enzyme variants, enhancing the overall throughput of enzyme engineering campaigns.
Notably, the researchers utilized augmented ridge regression models built on their extensive dataset, enabling them to predict variants likely to excel at producing specific small molecules. For example, the ML-optimized mutants demonstrated 1.6 to 42-fold improvements for the synthaesis of nine different pharmaceuticals, highlighting the framework's effectiveness.
Importantly, the ML-guided framework allows for future iterations, as every iteration provides more data, which can be leveraged to refine predictions for subsequent rounds of enzyme engineering. Overall, the approach not only accelerates the pace of enzyme optimization but also expands the potential for creating specialized biocatalysts across various domains.
The versatility of the McbA enzyme positions it as a promising starting point for engineering highly-efficient catalysts for pharmaceutical applications, potentially lowering costs and increasing the sustainability of production.
Moving forward, this machine learning-enhanced method could play a significant role, especially as researchers seek to tackle the challenge of broader and more complex chemical transformations.