Protein Fold Recognition Based on Auto-Weighted Multi-View Graph Embedding Learning Model.

Area of Science:

• Computational Biology
• Bioinformatics

Background:

• Protein fold recognition is crucial for protein structure prediction and drug design.
• Existing methods for extracting discriminative features from protein sequences have limitations, especially in ensemble approaches.
• Combining diverse features effectively for improved predictive performance remains a significant challenge.

Purpose of the Study:

• To develop novel algorithms for enhanced protein fold recognition.
• To address the challenges in ensemble methods for combining various protein features.
• To improve the accuracy and efficiency of predicting protein folds.

Main Methods:

• Proposed AWMG (Auto-Weight Multi-view Graph) algorithm using a multi-view learning framework with evolutionary and retrieval information.
• AWMG calculates auto-weights for each view, constructs a latent subspace of common information, and employs marginalized constraints to improve fold separation.
• Developed EMfold, an ensemble method combining AWMG with DeepSS (a template-based method using SPARKS-X and DeepFR), integrating template assignment and machine learning classification.

Main Results:

• AWMG demonstrated improved predictive performance by effectively utilizing multi-view information and marginalized constraints.
• EMfold successfully integrated the strengths of template-based and machine learning approaches.
• Both AWMG and EMfold outperformed several state-of-the-art methods on the LE and YK benchmark datasets.

Conclusions:

• AWMG and EMfold represent significant advancements in protein fold recognition.
• The proposed methods offer valuable tools for protein structure prediction and drug design.
• The integration of multi-view learning and ensemble strategies shows great promise for future bioinformatics research.