GDFuzz3D

GDFuzz3D reconstructs protein three-dimensional (3D) structures from predicted contact maps to generate accurate 3D models despite high false-contact rates typical of de novo contact predictions.

Key Features:

• Novel Distance Function: Employs a fuzzy residue proximity graph to define a distance function that predicts 2D distance maps from predicted contact maps and mitigates high fractions of false contacts from de novo predictions.
• Multi-Step Protocol: Transforms predicted 2D distance maps into coarse 3D models using Multi-Dimensional Scaling (MDS) and refines these models through modeling programs to obtain all-atom representations.
• Performance Superiority: Benchmarked against contact maps predicted by MULTICOM (CASP10) and compared to FT-COMAR, it consistently produces more accurate 3D models across sensitivity levels 60–84%, with an average improvement of 4.87 Å in root-mean-square deviation (RMSD).

Scientific Applications:

• Protein structure reconstruction from contact maps: Enables generation of 3D models from predicted contact maps that contain substantial numbers of false contacts.
• Structural biology and interaction analysis: Provides improved structural models to support studies of protein function and protein–protein interactions.

Methodology:

Uses a fuzzy residue proximity graph to compute a novel distance function that converts predicted contact maps into 2D distance maps, applies Multi-Dimensional Scaling (MDS) to obtain a coarse 3D model, and refines the model with modeling programs to produce an all-atom representation.