Katalyst: Knowledge-Guided Semantic Alignment via Contrastive Learning for Enzyme Turnover Prediction

The turnover number (kcat) is crucial for understanding enzyme kinetics, yet its experimental determination remains costly and low-throughput. Existing deep learning approaches often rely on single-substrate assumptions and overlook full reaction contexts or language-informed semantic representations, limiting their accuracy and generalization across diverse reactions.

To address these challenges, we propose Katalyst, a knowledge-guided semantic alignment framework for kcat prediction based on contrastive learning. Katalyst integrates enzyme sequences, protein textual descriptions, substrates and products into a shared latent space using multi-channel encoders and contrastive alignment. An interaction-aware fusion module is designed to capture complementary cross-modal information. Experimental results demonstrate that Katalyst achieved the best overall performance, with the RMSE of 0.9880, R2 of 0.5829, and SRCC of 0.7603. Compared with the State-Of-The-Art (SOTA) baselines, Katalyst reduced RMSE by 3.01% and improved R2 and SRCC by 4.72% and 2.31%, respectively. Moreover, ablation studies and evaluations across sequence scales further confirmed the effectiveness of semantic alignment and multimodal feature fusion. Overall, Katalyst provides a robust and accurate framework for kcat prediction.

The code and data are available at https://github.com/liruilin-SUAT/Katalyst.

Supplementary data are available at Bioinformatics online.