GTT-EC: a hierarchical graph transformer framework for enzyme function prediction via functional residue modeling

Current enzyme commission (EC) number prediction methods often rely on functional residue annotations and lack robustness. Here, we propose GTT-EC, a hierarchical graph transformer framework for EC number prediction. Without requiring explicit residue annotations, GTT-EC constructs protein graphs to integrate sequential and structural features to learn discriminative representations. Furthermore, it utilizes attention mechanisms to capture long-range interactions and dynamically focus on key functional regions. Experiments demonstrate that GTT-EC outperforms state-of-the-art methods in both accuracy and robustness across varying sequence similarities. Importantly, it effectively identifies functional regions without prior site inputs, demonstrating broad potential for downstream applications.