Comprehensive assessment of protein loop modeling programs on large-scale datasets: prediction accuracy and...

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein modeling

Background:

• Protein loops are crucial for protein dynamics and biological functions.
• Numerous computational loop modeling methods exist, but their comparative strengths and weaknesses are not well-understood.

Purpose of the Study:

• To systematically evaluate the accuracy and efficiency of 13 common loop modeling approaches.
• To provide insights for selecting appropriate methods for specific loop modeling tasks.

Main Methods:

• Construction of two high-quality datasets: General and CASP.
• Systematic evaluation of 13 loop modeling approaches based on loop length, protein class, and residue type.

Main Results:

• FREAD (knowledge-based) generally outperformed other methods but struggled with loops >15 (CASP) and >30 (General) residues.
• Rosetta NGK (ab initio) showed high accuracy for short loops (4-8 residues) and the best success rate on the CASP dataset.
• AlphaFold2 and RoseTTAFold require significant resources but show potential for modeling longer loops (>16 on CASP, >30 on General).

Conclusions:

• Method selection for protein loop modeling depends on loop length and dataset characteristics.
• FREAD, Rosetta NGK, AlphaFold2, and RoseTTAFold offer distinct advantages for different loop modeling scenarios.
• This comparative analysis aids in advancing computational protein structure prediction.