NanoNet

NanoNet predicts 3D coordinates of backbone and Cβ atoms for entire VH domains from amino acid sequences to enable structural modeling and epitope analysis of nanobodies and antibody variable domains.

Key Features:

• Deep Learning Architecture: An end-to-end deep learning model directly converts amino acid sequences into 3D coordinates of the backbone and Cβ atoms of the entire VH domain.
• Structural Resolution: Outputs explicit coordinates for backbone and Cβ atoms, covering variable domains and their Complementarity Determining Regions (CDRs).
• High Accuracy: For a nanobody (Nb) test set, average RMSD values are 3.16Å for CDR3, 2.65Å for CDR1, and 1.73Å for CDR2; for antibody VH domains, RMSDs are 2.38Å for CDR3, 0.89Å for CDR1, and 0.96Å for CDR2.
• Efficiency: Capable of generating approximately one million nanobody structures in under four hours on a standard CPU.

Scientific Applications:

• Therapeutic antibody development: Provides structural models to support design and optimization of antibodies and single-domain camelid VHH nanobodies.
• Nanobody engineering: Supplies atomic-coordinate models of VHH variable domains to guide sequence-to-structure analysis and engineering of stability and affinity.
• Epitope mapping: Enables molecular-level analysis of antigen–antibody interactions through predicted VH/VHH structures.

Methodology:

NanoNet uses a deep learning model trained to predict 3D nanobody and VH-domain structures from amino acid sequences, focusing on variable domains that share a common fold but differ in their CDRs.