Computational nanobody design using graph neural networks and Metropolis Monte Carlo sampling

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Nanobodies offer therapeutic advantages like stability and low immunogenicity.
• Experimental nanobody screening and optimization are resource-intensive due to extensive variant possibilities.

Purpose of the Study:

• To establish a computational framework for efficient nanobody design.
• To accelerate the development of novel protein therapeutics.

Main Methods:

• Integrated graph neural networks (GNNs) with the Monte Carlo Metropolis algorithm.
• Developed AiPPA, a GNN model to predict protein-protein binding free energy (BFE) without complex structures.
• Employed Metropolis importance sampling for designing low-BFE nanobodies from non-affinity templates.

Main Results:

• Achieved a Pearson correlation of 0.62 for BFE prediction using the AiPPA model.
• Successfully designed two high-affinity nanobodies against the TL1A antigen.
• Demonstrated a physics-informed deep learning method for computational nanobody design.

Conclusions:

• The developed computational approach significantly enhances nanobody design efficiency.
• This method offers a novel strategy for developing protein therapeutics.
• The integration of GNNs and Monte Carlo methods represents a breakthrough in computational protein design.