eThread: a highly optimized machine learning-based approach to meta-threading and the modeling of protein tertiary...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Template-based modeling is the leading method for protein structure prediction.
• Current methods struggle with a significant fraction of protein targets, necessitating new algorithms.

Purpose of the Study:

• To develop and benchmark eThread, a novel meta-threading procedure for accurate protein structure modeling.
• To enhance template identification, target-to-template alignment, and tertiary structure prediction.

Main Methods:

• eThread integrates ten state-of-the-art threading algorithms with machine learning for automated modeling.
• Utilizes Modeller and TASSER-Lite for tertiary structure prediction.
• Incorporates eContact for contact prediction and eRank for model ranking and quality assessment.

Main Results:

• eThread generates fold-level correct models for over 80% of targets, excluding closely related templates.
• 40-50% of predicted models exhibit high quality, accurate at the family level.
• Benchmarking demonstrates eThread's superior performance compared to alternative methods.

Conclusions:

• eThread represents a significant advancement in template-based protein structure prediction.
• The developed tools (eContact, eRank) enhance model quality assessment and confidence estimation.
• Further improvements are possible, with potential to refine upper bounds for this approach.