Performing protein fold recognition by exploiting a stack convolutional neural network with the attention mechanism.

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Bioinformatics

Background:

• Protein fold recognition is essential for predicting protein structure and function.
• Effective feature extraction from protein sequences is key to improving fold recognition accuracy.
• Current methods require robust feature extractors and metric functions for fold-specific patterns.

Purpose of the Study:

• To develop an effective sequence-based approach for protein fold type identification.
• To enhance protein fold recognition by extracting subtle patterns from residue-residue contact maps.

Main Methods:

• Proposed RattnetFold, employing a stacked convolutional neural network with an attention mechanism.
• Utilized RattnetFold as a feature extractor for protein residue-residue contact maps.
• Introduced RattnetFoldPro, which uses metric learning to project extracted features into a discriminative subspace.

Main Results:

• RattnetFold and RattnetFoldPro demonstrated improved performance in protein fold recognition.
• The methods efficiently learned subtle patterns within residue-residue contact maps using convolutional neural networks.
• Benchmarking experiments validated the effectiveness of the proposed approaches.

Conclusions:

• RattnetFold and RattnetFoldPro offer a powerful sequence-based strategy for protein fold recognition.
• The integration of attention mechanisms and metric learning significantly enhances prediction accuracy.
• An online web server and datasets are available for public use.