CASP10-BCL::Fold efficiently samples topologies of large proteins

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• BCL::Fold is a protein structure prediction tool that assembles tertiary structures from secondary structure elements (SSEs).
• The method omits flexible loop regions early in the process, enabling conformational sampling for large, complex proteins.

Purpose of the Study:

• To evaluate BCL::Fold's performance in the CASP10 competition for free modeling (FM) and template-based modeling (TBM) targets.
• To analyze the CASP10 prediction pipeline to identify areas for improvement in BCL::Fold for CASP11, focusing on topology sampling, model selection, and loop/side-chain addition.

Main Methods:

• Testing BCL::Fold on CASP10 free modeling and template-based modeling targets.
• Analyzing prediction pipeline stages: de novo topology sampling, scoring/clustering for native-like model identification, and loop/side-chain refinement.
• Evaluating model quality using GDT_TS and topology scores.

Main Results:

• BCL::Fold sampled topologies with GDT_TS > 33% for 12 of 18 targets and topology scores > 0.8 for 11 of 18 targets.
• Significant challenges were identified in the clustering and loop generation stages.
• Model refinement issues were observed for beta-strand proteins, and non-natural topologies requiring loop passage through the protein core were frequently sampled.

Conclusions:

• BCL::Fold demonstrates capability in sampling diverse protein topologies, particularly for complex structures.
• Improvements are needed in clustering algorithms for accurate model selection and in loop modeling techniques.
• Addressing refinement issues, especially for beta-strand proteins, and non-natural topology generation is crucial for enhancing prediction accuracy.