Protein secondary structure prediction using neural networks and deep learning: A review

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence

Background:

• Protein secondary structure prediction is crucial in bioinformatics.
• Over fifty years of research have yielded numerous prediction methods.
• Artificial neural networks (ANNs) are increasingly popular for this task.

Purpose of the Study:

• To review and synthesize key ANN-based methods for in silico protein secondary structure prediction.
• To provide a consolidated resource for researchers entering the field.
• To guide future research for enhanced prediction accuracy.

Main Methods:

• Review of literature focusing on ANN architectures and feature spaces for protein secondary structure prediction.
• Analysis of methods demonstrating significant improvements in prediction accuracy.
• Inclusion of introductory material on protein secondary structure and ANNs.

Main Results:

• ANNs with larger feature scopes and complex architectures yield better prediction results.
• Current prediction accuracy stands at approximately 84%.
• The theoretical prediction accuracy limit of 88% remains unreached.

Conclusions:

• Further research is essential to bridge the gap between current and theoretical prediction accuracy.
• Optimizing ANN architectures and feature spaces is key to advancing the field.
• This review serves as a foundation for developing next-generation prediction methods.