Accurate prediction of protein-nucleic acid complexes using RoseTTAFoldNA

Area of Science:

• Structural biology
• Computational biology
• Biochemistry

Background:

• Protein-nucleic acid complexes are vital for biological processes.
• Predicting the structures of these complexes, especially novel ones, remains challenging.
• Existing protein structure prediction methods do not adequately address nucleic acid interactions.

Purpose of the Study:

• To extend machine learning-based protein structure prediction to protein-nucleic acid complexes.
• To develop a computational tool for modeling protein-DNA and protein-RNA structures.
• To improve the accuracy and speed of predicting these complex structures.

Main Methods:

• Adapted the RoseTTAFold deep learning framework.
• Developed a single neural network, RoseTTAFoldNA, trained for protein-nucleic acid complex prediction.
• Integrated confidence estimation into the structure models.

Main Results:

• RoseTTAFoldNA rapidly generates 3D structure models for protein-DNA and protein-RNA complexes.
• Confident predictions from RoseTTAFoldNA demonstrate significantly higher accuracy than existing state-of-the-art methods.
• The tool provides reliable confidence estimates for the predicted structures.

Conclusions:

• RoseTTAFoldNA is a powerful new tool for modeling protein-nucleic acid complexes.
• The approach is broadly applicable to understanding naturally occurring complexes.
• It facilitates the design of novel proteins with specific RNA and DNA-binding capabilities.