Improved Functional Classification of Hydrolases through Pairwise Structural Similarity of Reaction Cores

Abstract

We report a systematic pipeline is for extracting the catalytically relevant reactive site in addition to the surrounding allosterically linked residue shells around the reaction site of the most diverse enzyme class - hydrolases, with known experimental structures. We first successfully extract 40196 such hydrolase reaction cores (RC) and collates them into a publicly accessible reaction core collection (RC-Hydrolase). We perform 128M pairwise shape comparison across RC-Hydrolase using a three-dimensional search engine and present 155,329 pair instances clustering them by 60% or higher similarities in a publicly available, visually interactive dataset. Robustness of defined RCs is shown to successfully capture experimentally known function-enhancing mutations distal to the active site in PETases. Allowing comparisons of enzyme reaction centers across functional spaces (ligands bound, EC classification numbers, and expression hosts) enables identification of enzyme backbones which can be minimally mutated to accommodate more than one type of catalytic activity thereby aiding rational design of multifunctional enzymes. We also demonstrate how such versatile enzyme backbones could be leveraged by the latest diffusion-based protein design models to design bespoke libraries of small molecule inhibitors, and structurally stable multifunctional enzyme pockets. With only sporadic successes in multifunctional enzyme design thus far, we provide strong structural priors for machine-learning-guided advanced enzyme engineering in the future.