OPUS-Rota3

OPUS-Rota3 predicts protein side-chain rotamers and refines side-chain dihedral-angle conformations using deep neural networks and ensemble rotamer libraries to improve accuracy in protein structural modeling.

Key Features:

• OPUS-RotaNN (Deep Neural Network Integration): OPUS-RotaNN predicts side-chain rotamers using deep learning rather than relying solely on traditional rotamer library sampling.
• Ensemble Rotamer Library: OPUS-Rota3 integrates results from various modeling methods into a combined rotamer library to provide diverse candidate conformations for sampling.
• Side-chain Dihedral-Angle Prediction with 20° Criterion: The framework evaluates side-chain dihedral-angle predictions using a 20° tolerance for accuracy assessment.
• Benchmark Performance on CAMEO-Hard61: On the native backbone test set CAMEO-Hard61, OPUS-Rota3 predicts 51.14% of side-chain dihedral angles correctly, compared with OSCAR-star 50.87%, SCWRL4 50.40%, and FASPR 49.85%.
• Benchmark Performance on DB379-ITASSER: On the non-native backbone test set DB379-ITASSER, OPUS-Rota3 achieves 52.49% accuracy, compared with OSCAR-star 48.95%, FASPR 48.69%, and SCWRL4 48.29%.

Scientific Applications:

• Protein Structure Prediction: Predicts side-chain dihedral angles to construct detailed and precise atomic models of protein structures.
• Comparative Structural Analysis and Benchmarking: Serves for comparative evaluation of side-chain modeling methods on native and non-native backbone datasets such as CAMEO-Hard61 and DB379-ITASSER.

Methodology:

OPUS-RotaNN predicts rotamers with deep neural networks; predicted rotamers are integrated into a combined rotamer library drawn from multiple methods; performance is validated on CAMEO-Hard61 and DB379-ITASSER using a 20° dihedral-angle tolerance.