Retrieving backbone string neighbors provides insights into structural modeling of membrane proteins

Area of Science:

• Structural biology
• Bioinformatics
• Computational biology

Background:

• Protein structure and function prediction are crucial for biological research.
• Identifying structural neighbors aids in modeling and understanding protein functions.
• Existing methods face challenges, especially for proteins with difficult structural determination.

Purpose of the Study:

• To introduce BS-align, a novel systematic method for retrieving protein structural neighbors.
• To utilize backbone strings for protein modeling and structure prediction.
• To demonstrate the efficacy of BS-align for modeling challenging membrane proteins.

Main Methods:

• Representing protein backbone conformation using backbone strings in Ramachandran space.
• Predicting backbone strings via knowledge-driven sequence alignment and element profile encoding.
• Aligning predicted backbone strings against a database to retrieve neighbors.
• Utilizing retrieved neighbors as templates for protein modeling.

Main Results:

• BS-align successfully retrieved backbone string neighbors close to native protein structures.
• Models of 10 membrane proteins were accurately predicted using BS-align.
• High TM-scores and low root mean square deviations confirmed model accuracy, despite low sequence identity.
• The backbone string system proved effective for modeling proteins with difficult structural determination.

Conclusions:

• The backbone string system offers a new paradigm for protein structure prediction from sequence.
• Backbone string similarity is a more informative metric than traditional fold classification for protein relationships.
• BS-align provides a robust approach for template-based protein modeling, particularly for challenging targets.