Distance-guided protein folding based on generalized descent direction.

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Accurate protein structure prediction is crucial for understanding biological function.
• Predicting inter-residue distances has significantly improved protein folding accuracy.
• Existing methods face challenges in exploring conformational space and optimizing potential energy landscapes.

Purpose of the Study:

• To introduce GDDfold, a new distance-guided protein folding algorithm.
• To enhance structural perturbation and potential minimization for protein folding.
• To evaluate GDDfold's performance against established methods and state-of-the-art approaches.

Main Methods:

• GDDfold employs a two-stage approach: a global stage for broad conformational exploration and a local stage for detailed optimization.
• The global stage utilizes evolutionary knowledge for random-based direction to overcome potential barriers.
• The local stage employs conjugate-based direction and a specialized search strategy for fine-tuning structures.

Main Results:

• GDDfold successfully folded 316 out of 347 proteins (TM-score ≥ 0.5) on a benchmark set.
• 65 proteins achieved TM-scores greater than 0.8, indicating high accuracy.
• Performance on CASP13 and CASP14 free modeling targets was comparable to leading methods like Quark and RaptorX.

Conclusions:

• GDDfold demonstrates superior performance in protein structure prediction compared to Rosetta-dist and L-BFGSfold.
• The algorithm's two-stage strategy effectively balances exploration and exploitation of conformational space.
• GDDfold represents a significant advancement in computational protein folding, achieving competitive results on challenging targets.