GDDfold

GDDfold predicts protein tertiary structures by leveraging inter-residue distance information and a generalized descent direction algorithm to explore and optimize conformational space.

Key Features:

• Distance-Guided Folding Algorithm: Uses inter-residue distance information to guide conformational changes during structure prediction.
• Generalized Descent Direction: Implements a generalized descent direction algorithm to drive potential minimization and folding optimization.
• Two-Stage Optimization Process: Combines a global stage using a random-based direction informed by evolutionary information for broad exploration and a local stage using a conjugate-based direction for refined exploitation in locally rugged potentials.
• Evolutionary Information Integration: Incorporates evolutionary-derived restraints to inform guidance of conformational sampling and barrier crossing.
• Potential Minimization: Applies potential minimization as an optimization objective during folding.
• Benchmark Performance: Evaluated on a benchmark of 347 proteins, achieving TM-score ≥ 0.5 for 316 proteins and TM-score > 0.8 for 65 proteins.
• Comparative Success: Demonstrated superior performance relative to Rosetta-dist and L-BFGSfold and shown competitive results against Quark, RaptorX, Rosetta, MULTICOM, and trRosetta on CASP 13 and 14 FM targets.

Scientific Applications:

• Protein structure prediction: Predicts tertiary structures from sequence-derived inter-residue distances.
• Functional inference: Supports interpretation of protein function through predicted structural models.
• Drug design: Provides structural models that can be used to inform small-molecule or biologics design efforts.
• Computational structural biology benchmarking: Serves as a method for comparative evaluation on CASP FM targets and curated protein benchmark sets.

Methodology:

Leverages inter-residue distance information with a generalized descent direction algorithm in a two-stage optimization: a global random-based direction informed by evolutionary information for broad exploration, followed by a local conjugate-based direction for refined exploitation and potential minimization.