.png){kind=icon}
Return
SEFP: Structure-Based Enzyme Function Prediction
.png){kind=icon}
PRE
AI
DOI：10.1109/TCBBIO.2025.3552846.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traditional biological experimental methods to determine enzyme properties are time-consuming and costly, leading to an increasing interest in computational models for enzyme function prediction. However, the existing computational methods are insufficient and inefficient to exploit enzyme structure. In this work, we introduce SEFP, a novel method leveraging enzyme point clouds for enzyme function prediction. The structure encoder of SEFP uses a tailored enzyme point cloud network to analyze the three-dimensional arrangement of atoms within the enzyme, integrating hierarchical residue global features through a residue feature adapter to extract detailed enzyme point features. Additionally, the Bio-BCS residue feature encoder extracts enzyme residue features with channel and spatial weights using a specially designed attention mechanism. Finally, SEFP fuses point and residue features to generate the final prediction results. Comparative evaluations show that SEFP outperforms various recent computational methods, demonstrating superior performance. On the RSCB enzyme structure dataset, SEFP achieves an f1-score of 95.85, outperforming two representative structure-based methods, EnzyNet and DeepFri. On the HECNet dataset, SEFP maintains its superiority over all comparison sequence-based methods, yielding an f1-score of 94.29. Ablation studies are conducted to confirm the effectiveness of individual modules within SEFP. These findings underscore the potential of SEFP for reliable and precise enzyme function prediction, offering advancements in bioinformatics and computational biology.
Keywords:
Enzyme function prediction
point cloud
structural feature learning
attention mechanism
Journal
I
IF:
3.4
Papers:
3.3K
Citations:
6.4K
Organization
Cited Papers
SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15-year Anniversary
OAAI
.png){kind=icon}
OAAIShareSaveMVDINET: A Novel Multi-Level Enzyme Function Predictor With Multi-View Deep Interactive Learning
PREAI
ShareSaveShareSave
ShareSaveShareSaveShareSaveShareSave
View more