A new artificial intelligence tool called EZSpecificity helps researchers predict how well an enzyme will interact with a specific substrate, significantly improving the ability to identify optimal enzyme-substrate pairs for applications in medicine, manufacturing, and chemical catalysis. Developed by a team led by Huimin Zhao, a professor of chemical and biomolecular engineering at the University of Illinois Urbana-Champaign, the tool uses advanced machine learning and a large, experimentally supported dataset to make accurate predictions. Enzymes are proteins that speed up chemical reactions by binding to specific molecules called substrates. The strength and precision of this binding—known as specificity—is crucial for efficient and targeted reactions. Unlike the traditional "lock and key" analogy, enzyme-substrate interactions are more dynamic, with enzymes changing shape upon substrate binding—a concept known as induced fit. Some enzymes are also promiscuous, capable of acting on multiple substrates, making prediction challenging. To address these complexities, Zhao’s team collaborated with Diwakar Shukla, another U. of I. professor, whose group performed extensive molecular docking simulations. These simulations modeled atomic-level interactions between enzymes and substrates across various enzyme classes, generating a comprehensive dataset that fills gaps in existing experimental data. Millions of docking calculations provided detailed insights into how enzymes adapt to different substrates. Using this expanded dataset, the team trained EZSpecificity, a machine learning model that analyzes an enzyme’s amino acid sequence and predicts which substrates will bind most effectively. The model was tested against ESP, the current leading specificity prediction tool, across four real-world scenarios. EZSpecificity outperformed ESP in every case. Experimental validation further confirmed its accuracy. In a test involving 8 halogenase enzymes—enzymes used to create bioactive compounds—and 78 substrates, EZSpecificity correctly identified the best substrate pair in 91.7% of top predictions, compared to just 58.3% for ESP. The researchers have made EZSpecificity freely available online with a user-friendly interface. Scientists can input an enzyme sequence and a substrate structure to receive predictions on binding compatibility. Zhao emphasized that while the tool is not universally applicable to all enzymes, it performs exceptionally well for specific classes, particularly those with limited prior characterization. Future work will focus on enhancing the model to predict enzyme selectivity—identifying whether an enzyme acts on a specific site of a substrate—which is vital for minimizing unwanted side reactions. The team also plans to continue refining EZSpecificity by incorporating additional experimental data.