UniCon3D

UniCon3D predicts de novo protein tertiary structures by performing stepwise, probabilistic conformational sampling using a united-residue generative model combined with physics-based and knowledge-based energy functions.

Key Features:

• Generative probabilistic model: Captures local structural preferences within both backbone and side chain conformational spaces using a united-residue representation.
• Stepwise conformational sampling: Performs conditional, sequential synthesis and assembly of foldon units rather than random sampling of the entire conformational space.
• Energy minimization: Minimizes a composite energy function that combines physics-based and knowledge-based components to favor lower-energy conformations.
• Contact integration: Supports modeling aided by predicted residue-residue contacts for guiding conformational sampling.

Scientific Applications:

• Lower-energy conformation benchmarking: Sampled lower-energy conformations with greater precision than traditional methods on a benchmark set of six proteins.
• CASP11 comparative performance: Performed comparably with the top five automated methods in the 11th Critical Assessment of Protein Structure Prediction (CASP11) across thirty challenging target domains.
• CASP10 difficult targets: Showed comparable performance on fifteen difficult targets from the 10th CASP experiment.
• Contact-aided modeling benchmark (CASP10): Outperformed two state-of-the-art approaches and a random-sampling baseline in contact-aided protein modeling across forty-five CASP10 targets.

Methodology:

Uses a united-residue generative probabilistic model to perform conditional, stepwise sampling by synthesizing and assembling foldon units and minimizes a composite energy function combining physics-based and knowledge-based components, with optional incorporation of predicted residue-residue contacts.