SSE-TSR: An Approach to Integrate Secondary Structure Elements Into Triangular Spatial Relationships for Protein...

Area of Science:

• Structural bioinformatics
• Computational biology
• Protein structure analysis

Background:

• Conventional protein structure comparison methods struggle to identify subtle similarities.
• Triangular Spatial Relationship (TSR) provides alignment-free geometric encoding but lacks secondary structure context.
• Secondary structure elements (SSEs) like helices, strands, and coils are crucial for protein function.

Purpose of the Study:

• To introduce SSE-TSR, an enhanced alignment-free method for protein structure representation.
• To integrate secondary structure information into TSR for a more comprehensive encoding.
• To evaluate SSE-TSR's performance on structural and functional classification tasks.

Main Methods:

• Developed SSE-TSR by enriching TSR keys with 18 helix-strand-coil combination labels derived from DSSP annotations.
• Mapped SSE-TSR keys into a sparse tensor representation to capture tertiary geometry and local secondary motifs.
• Utilized a 3D convolutional neural network for evaluating SSE-TSR on structural (CATH, SCOP) and functional datasets.

Main Results:

• SSE-TSR significantly improved accuracy on structure-based tasks: CATH-based (96.00% to 98.33%) and SCOP-based (95.46% to 99.00%).
• Modest yet consistent accuracy gains were observed for functional tasks.
• SSE-TSR demonstrated competitive accuracy compared to Foldseek and enabled memory-efficient handling of large datasets.

Conclusions:

• SSE-TSR is a scalable, interpretable, and robust method for protein structure analysis.
• The integration of secondary structure context enhances the representation of protein geometry.
• SSE-TSR offers practical advantages for large-scale bioinformatics analyses and improves protein classification accuracy.