IPTDFold

IPTDFold performs de novo protein structure prediction using a closed-loop continuous dihedral angle optimization strategy that integrates iterative partition sampling, topology adjustment, and residue-level distance deviation optimization to improve conformational sampling and structural accuracy.

Key Features:

• Iterative Partition Sampling: Employs local dihedral angle crossover and mutation operators to explore conformational space and enable information exchange within a population.
• Topology Adjustment: Constructs a dihedral angle rotation model for loop regions with partial inter-residue distance constraints and uses a differential evolution algorithm to identify rotation angles that satisfy those constraints.
• Residue-Level Distance Deviation Optimization: Evaluates residue distance deviations by comparing predicted distances to current conformations and optimizes dihedral angles with biased probability to minimize deviations.

Scientific Applications:

• Benchmark Evaluation: Validated on 462 benchmark proteins to assess prediction accuracy.
• CASP Free-modeling Targets: Applied to 24 CASP13 and 20 CASP14 free-modeling (FM) targets.
• Comparative Performance: Outperforms Rosetta_D in prediction accuracy and maintains robustness across varying protein lengths compared with methods such as trRosetta when using the same FastRelax protocol.

Methodology:

Iteratively: local dihedral angle crossover and mutation for conformational exploration and information exchange; differential evolution algorithm to optimize dihedral rotation angles in loop regions under partial inter-residue distance constraints; biased probability optimization of dihedral angles to minimize residue-level distance deviations.