RNA sequence design and protein-DNA specificity prediction with NA-MPNN.

Area of Science:

• Computational biology
• Structural biology
• Machine learning

Background:

• RNA sequence design and protein-DNA binding specificity prediction are inverse-folding problems.
• Existing methods are task-specific, limiting unified deep learning applications.
• A single model could leverage larger datasets and offer broader applicability.

Purpose of the Study:

• To introduce a unified deep learning model for nucleic acid inverse folding.
• To develop a message-passing neural network for biopolymer graph representation.
• To improve RNA sequence design and protein-DNA binding specificity prediction.

Main Methods:

• Developed Nucleic Acid MPNN (NA-MPNN), a message-passing neural network.
• Treated proteins, DNA, and RNA within a unified biopolymer graph representation.
• Applied NA-MPNN to RNA sequence design and protein-DNA specificity prediction tasks.

Main Results:

• NA-MPNN outperforms previous methods on RNA sequence design.
• NA-MPNN achieves superior performance in fixed-dock protein-DNA specificity prediction.
• Demonstrated the model's effectiveness across different nucleic acid and protein-DNA interaction tasks.

Conclusions:

• NA-MPNN provides a unified deep learning framework for nucleic acid inverse folding.
• The model shows significant improvements over existing methods.
• NA-MPNN is broadly applicable for de novo RNA structure design and DNA-binding specificity prediction.