ML-Guided Enzyme Engineering Advances Green Chemistry

Amide bonds — a type of chemical bond formed between a carbon atom and a nitrogen atom — are a fundamental component of many natural and synthetic materials including proteins, pharmaceuticals, synthetic materials, and everyday products such as agrochemicals, fragrances, and flavors.

Recent work from Northwestern Engineering’s Ashty Karim and Stanford University’s Michael Jewett take the creation of amide bonds to a new level. 

Karim and his colleagues developed a platform to engineer enzymes responsible for the formation of amide bonds as an effort toward developing biocatalysts for green chemistry. The work overcomes current conventional approaches to enzyme engineering, which are limited across multiple dimensions:

• Small functional datasets lead to missed interactions among different amino acid residues within a protein.
• Low-throughput screening strategies have difficulty in rapidly generating large functional datasets which machine learning (ML) models rely on.
• Selection methods focus on evolving “winning” enzymes for a single transformation, limiting the collection of sequence-function relationships, both positive and negative, for further engineering of similar reactions.

 “This work addresses these challenges with our new cell-free approach,” Karim said. “Engineered enzymes are poised to have transformative impacts on the bioeconomy across numerous applications in energy, materials, and medicine.”

Karim is an assistant professor in the McCormick School of Engineering’s Department of Chemical and Biological Engineering and member of Northwestern’s Center for Synthetic Biology. He is a co-corresponding author of the work, “Accelerated Enzyme Engineering by Machine-Learning Guided Cell-Free Expression,” published January 20 in the journal Nature Communications. Jewett is an adjunct professor at the McCormick School of Engineering.

Karim and his team created a novel high-throughput, cell-free, and ML-guided platform to iteratively explore sequence-fitness landscapes across multiple regions of chemical space. They applied the platform by assessing 1,217 mutants of an amide synthetase — an enzyme that catalyzes two adjoining molecules — called McbA, from the Marinactinospora thermotolerans bacterium, in 10,953 unique reactions to map the functionality of McbA. They used the resulting data to train a ML model to predict multiple amide synthetase variants capable of making nine small molecule pharmaceuticals.

“Our approach enabled us to engineer McbA for six compounds simultaneously,” Karim said. “In all cases, newly generated enzyme variants demonstrated improved amide bond formation. Taken together, our work highlights both the versatility of McbA to be directed to catalyze many unique reactions of interest and, importantly, a new ability to iteratively explore protein sequence space to rapidly build specialized biocatalysts in parallel.”

This work builds off the extensive research in protein engineering as well as in cell-free synthetic biology at Northwestern. It combines the customizable reaction environment of cell-free systems that enables high-throughput experimentation and ML with sequenced defined mutational libraries for screening unique protein variants.

However, there is more work to be done.

“New and powerful artificial intelligence approaches are coming out rapidly,” Karim said. “We would like to take advantage of these new methods to make our workflows even more capable of creating new-to-nature proteins. In addition, we are trying to leverage our approach to engineer enzymes not just for activity but also for stability and industrial use.”