LOMETS3: integrating deep learning and profile alignment for advanced protein template recognition and function...

Area of Science:

• Computational biology
• Structural bioinformatics
• Deep learning applications

Background:

• Protein structure prediction is crucial for understanding biological function.
• Template-based modeling is a key approach, but challenges remain for complex proteins.
• Deep learning has shown promise in improving prediction accuracy.

Purpose of the Study:

• To introduce LOMETS3, an advanced meta-server for template-based protein structure prediction and function annotation.
• To extend protein structure prediction capabilities to multi-domain proteins.
• To improve the accuracy and scope of protein modeling using deep learning.

Main Methods:

• Integration of novel deep learning threading methods into the LOMETS3 meta-server.
• Development of techniques to handle multi-domain proteins and construct full-length models.
• Application of gradient-based optimization (L-BFGS) for model assembly.
• Four-step pipeline: domain boundary prediction, template identification, model assembly, and function prediction.

Main Results:

• LOMETS3 demonstrates significantly improved template recognition and function prediction accuracy compared to predecessors and state-of-the-art methods.
• Exceptional performance observed for challenging targets lacking homologous templates in the Protein Data Bank (PDB).
• Successful handling of multi-domain proteins and construction of full-length models validated in large-scale benchmarks and CASP14.

Conclusions:

• LOMETS3 represents a significant advancement in template-based protein modeling.
• The tool enhances the ability to predict protein structure and function, particularly for difficult targets.
• LOMETS3 is expected to benefit the broader biomedical community by improving protein modeling capabilities.