RC-GNN: A predictive model of enzyme-reaction pairs

Uncharacterized functions of enzymes represent untapped opportunity to develop therapeutics, unlock the sustainable synthesis of materials, and understand the evolution of life-sustaining metabolic networks. Enzymes and de novo reactions (i.e., non-native, promiscuous reactions), generated by protein language models and computer-aided synthesis tools, respectively, make up a large part of this opportunity. Given the technical complexity of high-throughput enzymatic activity screens, predictive models are needed that can pre-screen de novo enzyme-reaction pairs in silico. We present Reaction-Center Graph Neural Network, (RC-GNN) a model capable of predicting whether an enzyme, represented by an amino acid sequence, can significantly catalyze a given reaction, represented by its full set of reactants and products. We explicitly evaluated RC-GNN's generalization to de novo queries. In the most difficult conditions tested, where difficulty is measured by the level of dissimilarity between training and test data points, the model achieves 78.0% and 94.8% accuracy when reaction and enzyme similarity were respectively controlled. The ability to successfully make predictions on enzymes and reactions distinct from those used during training make RC-GNN especially useful for both metabolic engineers and evolutionary biologists who need to reason about uncharacterized enzymatic reactions.