The trRosetta server for fast and accurate protein structure prediction.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein structure prediction is crucial for understanding biological function.
• Accurate and rapid prediction methods are needed for large-scale biological studies.
• Existing methods may lack speed or accuracy for de novo predictions.

Purpose of the Study:

• To introduce the trRosetta server for fast and accurate protein structure prediction.
• To provide a web-based platform and downloadable package for protein structure modeling.
• To demonstrate the utility of deep learning combined with Rosetta for structure prediction.

Main Methods:

• Utilizes a deep neural network to predict inter-residue geometries (distance, orientation) from amino acid sequences.
• Employs predicted geometries as restraints for structure prediction via Rosetta's energy minimization.
• Integrates optional homologous templates to enhance prediction accuracy.

Main Results:

• Achieved rapid (approx. 1 hour for 300 amino acids) and accurate de novo protein structure prediction.
• Demonstrated high accuracy on Pfam families with previously unknown structures.
• The server and open-source package are publicly available for broad use.

Conclusions:

• trRosetta offers a significant advancement in speed and accuracy for protein structure prediction.
• The combination of deep learning and Rosetta provides a powerful framework for modeling.
• The availability of the server and package facilitates large-scale structural biology research.