DeepMVD: A Novel Multiview Dynamic Feature Fusion Model for Accurate Protein Function Prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Proteins are essential macromolecules involved in numerous biological processes.
• Accurate protein function prediction is crucial for understanding biological systems.
• Existing methods often fail to fully utilize multilevel attribute features from protein sequences.

Purpose of the Study:

• To develop a novel deep learning model, DeepMVD, for enhanced protein function prediction.
• To effectively integrate multilevel attribute features from protein sequences.
• To improve the accuracy of protein function prediction using sequence data.

Main Methods:

• Proposed DeepMVD, a deep learning model utilizing dynamic fusion of multiview features.
• Employed specialized modules for extracting unique features from each data view.
• Utilized an adaptive fusion mechanism for optimal integration of extracted features.

Main Results:

• DeepMVD demonstrated significant performance improvements over state-of-the-art models on the CAFA4 dataset.
• Achieved highest Fmax scores for Biological Process (BP) (0.523), Molecular Function (MF) (0.712), and Cellular Component (CC) (0.740) terminology.
• Ablation studies confirmed the robustness and effectiveness of the DeepMVD model.

Conclusions:

• DeepMVD offers a powerful new approach for protein function prediction by effectively leveraging multilevel sequence features.
• The model's ability to dynamically fuse multiview features leads to superior prediction accuracy.
• The findings provide a valuable tool for advancing research in bioinformatics and computational biology.