BASILISK

BASILISK models the continuous, generative probabilistic conformational space of protein amino acid side chains, enabling backbone-conditional sampling and improved side-chain representation for protein design, docking, and high-resolution structure prediction.

Key Features:

• Continuous Space Modeling: Represents amino acid side chain conformations in continuous space rather than as discrete rotamers.
• Conditional Sampling: Performs sampling conditional on detailed backbone conformation to maintain coherence between side chain and backbone geometry.
• Integration with Physical Force Fields: Facilitates integration with physical force fields to enable rigorous, unbiased sampling compared with discrete rotamer approaches.
• Pseudo-Energy Term for Improved Prediction: Provides a pseudo-energy term that can be used to enhance side-chain prediction accuracy relative to traditional methods.

Scientific Applications:

• Protein Design: Enables more accurate side-chain modeling to inform sequence and structural design decisions.
• Docking Studies: Improves representation of potential binding sites through detailed continuous side-chain sampling.
• Structure Prediction: Enhances high-resolution structure prediction by supplying a rigorous probabilistic description of side-chain conformational ensembles.

Methodology:

BASILISK employs a generative probabilistic model to represent side-chain conformational space, performs continuous sampling conditional on backbone conformation, and compares sampled distributions against rotamer libraries.