CGLFold

CGLFold predicts de novo protein tertiary structures by combining global fragment-based exploration with loop-specific perturbation sampling to improve conformational accuracy.

Key Features:

• Global Exploration Phase: Employs fragment recombination and assembly to navigate conformational space and generate native-like topologies.
• Loop Perturbation Phase: Uses a loop-specific local perturbation model to refine loop conformations and increase structural diversity.
• Differential Evolution Optimization: Solves the local perturbation model with differential evolution to enhance conformational updates.
• Cooperative Strategy: Integrates global exploration with local exploitation to synergistically refine predicted structures.
• Contact Information Utilization: Uses filtered contact information to guide the conformation selection model and direct sampling toward more accurate representations.

Scientific Applications:

• Benchmark proteins: Evaluated on 145 standard test proteins, achieving template modeling scores (TM-score) ≥ 0.5 for 95 of those proteins.
• CASP free modeling targets: Achieved TM-score ≥ 0.5 for 7 free modeling targets in CASP13 and 9 free modeling targets in CASP12.
• Conformational accuracy improvement: Demonstrates improved conformational accuracy and increased success rates through its combined global and loop-specific sampling approach.

Methodology:

Two-phase computational approach: Phase 1 performs global exploration via fragment recombination and assembly; Phase 2 applies a loop-specific local perturbation model optimized by differential evolution.