Atomic context-conditioned protein sequence design using LigandMPNN.

Area of Science:

• Computational biology
• Protein engineering
• Deep learning

Background:

• Current deep learning methods struggle to model nonprotein components in biomolecular systems.
• Designing proteins that bind small molecules, nucleotides, and metals is crucial for various applications.

Purpose of the Study:

• To introduce LigandMPNN, a novel deep learning-based protein sequence design method.
• To enable explicit modeling of nonprotein components (ligands) in protein design.

Main Methods:

• Developed LigandMPNN, a graph neural network architecture.
• Trained and evaluated LigandMPNN on protein sequence recovery tasks involving small molecules, nucleotides, and metals.
• Compared LigandMPNN performance against Rosetta and ProteinMPNN.

Main Results:

• LigandMPNN significantly outperformed Rosetta and ProteinMPNN in native backbone sequence recovery for residues interacting with small molecules (63.3% vs. 50.4%/50.5%), nucleotides (50.5% vs. 35.2%/34.0%), and metals (77.5% vs. 36.0%/40.6%).
• LigandMPNN generates sequences and sidechain conformations for detailed binding interaction analysis.
• Over 100 experimentally validated small-molecule and DNA-binding proteins were designed using LigandMPNN, achieving high affinity and structural accuracy.

Conclusions:

• LigandMPNN represents a significant advancement in protein sequence design, particularly for systems involving nonprotein components.
• The method enables the design of novel binding proteins, sensors, and enzymes with high specificity and affinity.
• LigandMPNN has demonstrated practical utility through successful experimental validation and significant improvements in binding affinity.