Abstract: We introduce HyperGenie, a method for predicting enzymatic gene essentiality by leveraging hypergraph neural networks and genome-scale metabolic models. In this approach, the $\textit{E. coli}$ metabolic network is represented as a hypergraph, with metabolites as nodes and reactions as hyperedges. HyperGenie achieves significant performance gains over existing methods, improving the area under the precision-recall curve (PR-AUC) by over 12\% for $\textit{E. coli}$ growing on glucose under aerobic conditions. It matches or surpasses the accuracy of flux balance analysis without assuming that knockout strains optimise for growth. Notably, HyperGenie maintains a PR-AUC above 90\% using only 40\% of the training data and exhibits robust performance across various nutrient conditions.