BioS2Net: Holistic Structural and Sequential Analysis of Biomolecules Using a Deep Neural Network

Area of Science:

• Structural bioinformatics
• Deep learning applications in biology
• Computational biology

Background:

• Manual classification of biomolecular structures lags behind discovery rates.
• Efficient, comprehensive tools are needed for biomolecular examination.
• Current methods face challenges in holistic feature characterization.

Purpose of the Study:

• To introduce a novel deep neural network for biomolecular analysis.
• To develop a tool integrating sequential and structural data.
• To enable efficient classification and feature extraction of biomolecules.

Main Methods:

• Proposed the Biological Sequence and Structure Network (BioS2Net).
• Utilized a deep neural network architecture with four components: sequence convolutional extractor, 3D structure extractor, 3D structure-aware sequence temporal network, and a fusion/classification network.
• Integrated sequential and structural information for comprehensive analysis.

Main Results:

• Achieved 95.4% mean class accuracy on the eDD protein fold classification dataset.
• Attained 76% mean class accuracy on the F184 dataset.
• Demonstrated performance comparable to state-of-the-art methods on the eDD dataset.

Conclusions:

• BioS2Net provides a novel, holistic tool for examining biomolecules.
• The network enables unified representation of proteins as feature vectors.
• It serves as a reliable tool for advanced protein analysis.