EvoDesign

EvoDesign generates de novo protein sequences for specified structural scaffolds by leveraging evolutionary profiles and a Monte Carlo search to optimize foldability and stability.

Key Features:

• Evolutionary Profile-Based Design: Constructs evolutionary profiles from homologous structure families within the Protein Data Bank (PDB) to guide sequence selection.
• Monte Carlo Search Algorithm: Employs an evolution-profile-based Monte Carlo search to explore sequence space and identify low-energy states while maintaining structural integrity.
• Structure and Sequence-Based Features: Integrates multiple sequence and structure-based features to evaluate foldability and design quality and to optimize physicochemical packing.
• Local Structural Attribute Prediction: Predicts secondary structure, torsion angles, and solvation properties using single-sequence neural network training to refine sequence motifs.
• Enhanced Foldability and Stability: Produces designed sequences with improved foldability and structural stability relative to traditional physics-based force field methods in large-scale tests.

Scientific Applications:

• Drug discovery: Design of novel protein scaffolds and binding interfaces for therapeutic targets.
• Enzyme design: Engineering of catalytic proteins with specified structural frameworks and optimized active-site packing.
• Synthetic biology: Creation of novel proteins for synthetic pathways and functional modules.
• Protein folding studies: Investigation of determinants of foldability and structural stability through designed sequence variants.
• Functional biomolecule development: Generation of proteins with tailored structure–function properties for research and application.

Methodology:

Constructs evolutionary profiles from homologous structure families in the PDB; applies an evolution-profile-based Monte Carlo search to sample sequence space and identify low-energy states; integrates multiple sequence- and structure-based features and predicts local attributes (secondary structure, torsion angles, solvation) using single-sequence neural network training; optimizes physicochemical packing and models atomic interactions to assess design quality.