Related Experiment Videos
Tri_PTM_DAAM_CNN: Tripterygium wilfordii Post-translational Modification with Density Adaptive Attention Mechanism
Yingyue Tang1, Baitong Chen2, Wenzheng Bao3
1School of Information Engineering, Yancheng Institute of Technology, Yancheng, 224051, China.
Current Computer-Aided Drug Design
|April 20, 2026
Summary
A new deep learning model, Tri_PTM_DAAM_CNN, efficiently predicts post-translational modification (PTM) sites in Tripterygium wilfordii proteins. This computational approach enhances accuracy and reduces time compared to traditional methods.
Area of Science:
- Computational biology
- Bioinformatics
- Deep learning applications in proteomics
Background:
- Identifying post-translational modification (PTM) sites in Tripterygium wilfordii proteins is crucial for understanding their pharmacological functions.
- Traditional methods for PTM site identification are laborious and time-consuming.
- Advancements in deep learning offer efficient computational solutions for complex biological datasets.
Purpose of the Study:
- To develop an accurate and efficient computational model for predicting PTM sites in Tripterygium wilfordii proteins.
- To leverage deep learning, specifically attention mechanisms and convolutional neural networks, for PTM site prediction.
- To address challenges posed by complex and imbalanced protein sequence datasets.
Main Methods:
- Developed the Tri_PTM_DAAM_CNN model, integrating the DAAM attention mechanism with convolutional neural networks (CNNs).
- Transformed one-dimensional protein sequence data into image representations for model input.
- Trained and evaluated the model on nine feature datasets, comparing its performance against established methods.
Main Results:
- The Tri_PTM_DAAM_CNN model achieved superior performance over classical models on BLOSUM Tripterygium wilfordii protein features.
- Achieved high accuracy (84.65%), Matthews Correlation Coefficient (0.6646), and F1 score (0.7824).
- Ablation studies confirmed the effectiveness of CNNs in local feature extraction and the DAAM mechanism's adaptability for PTM prediction.
Conclusions:
- The Tri_PTM_DAAM_CNN model successfully predicts PTM sites by dynamically integrating local and global protein sequence information.
- The model's architecture, combining CNNs with attention mechanisms, effectively extracts relevant features while filtering redundant data.
- The study highlights the importance of multiparameter tuning for assessing model stability in PTM site prediction.
