DNN-Dom: predicting protein domain boundary from sequence alone by deep neural network

Area of Science:

• Computational biology
• Bioinformatics
• Structural bioinformatics

Background:

• Accurate protein domain boundary delineation is crucial for protein engineering and structure prediction.
• Existing machine learning methods often neglect long-range residue interactions, limiting prediction accuracy.
• Simultaneously modeling local and global interactions remains a challenge in domain boundary prediction.

Purpose of the Study:

• To develop an advanced method for accurate protein domain boundary prediction.
• To integrate local and non-local residue interactions for enhanced prediction performance.
• To provide a robust tool for assisting protein structure and function prediction.

Main Methods:

• A hybrid deep learning approach combining Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU).
• Feature fusion to capture both local and non-local residue interactions.
• Balanced Random Forest for classification to handle imbalanced datasets and high-dimensional features.

Main Results:

• The proposed DNN-Dom approach demonstrates superior performance compared to existing machine learning-based methods.
• Effective capture and fusion of local and non-local interactions contribute to improved prediction accuracy.
• The method successfully addresses challenges posed by imbalanced samples and high-dimensional deep features.

Conclusions:

• DNN-Dom offers a significant advancement in protein domain boundary prediction.
• The hybrid deep learning model provides a powerful tool for computational biology research.
• DNN-Dom is expected to aid in protein structure and function prediction, advancing the field.