AI-Powered ESP Model Sheds Light on Enzyme Functions

Enzymes play a crucial role in chemical reactions, but determining which molecular building blocks they use to assemble target molecules has remained challenging. Recently, an international team, including bioinformaticians from Heinrich Heine University Düsseldorf (HHU), made progress in predicting enzyme-substrate compatibility using AI. The team’s findings have been published in Nature Communications.

Despite identifying the genes encoding certain enzymes, understanding their exact functions remains unknown in over 99% of cases. Notably, experimental characterizations to determine enzyme converts’ starting and end molecules can be time-consuming.

The team, led by Professor Dr. Martin Lercher from HHU’s Computational Cell Biology research group, developed an AI-based method called ESP (Enzyme Substrate Prediction) to predict whether an enzyme can work with a specific molecule as a substrate for catalysis.

The ESP model can work with any combination of an enzyme and over 1,000 different substrates, unlike previous models limited to specific enzymes and closely related variants. Lead author and Ph.D. student Alexander Kroll designed a Deep Learning model using numerical vectors encoding information about enzymes and substrates.

To train the model, the team used approximately 18,000 experimentally validated enzyme-substrate pairs as input. After training, they tested their model on an independent dataset where the researchers already knew the correct answers. The model reflected 91% accuracy in predicting which substrates match with which enzymes.

Knowing which substances can be converted by enzymes is critical for developing new drugs. By narrowing down potential enzyme-substrate pairs to the most promising ones, this method streamlines the enzymatic production of various products.

The ESP model can also assist in developing improved cell metabolism simulation models, enhance our understanding of different organisms’ physiology, from bacteria to humans, and provide valuable insights for fundamental and applied scientific research. By leveraging AI and Deep Learning techniques, the ESP method helps overcome previous limitations, offering researchers and industries a powerful tool to accelerate their discoveries and optimize biocatalytic processes.