HyperMPNN-A general strategy to design thermostable proteins learned from hyperthermophiles

Area of Science:

• Protein engineering
• Computational biology
• Biophysics

Background:

• Protein stability is crucial for therapeutic and biotechnological applications.
• Current deep learning models like ProteinMPNN are limited by the marginal stability of their training data.
• Natural proteins from hyperthermophiles possess unique amino acid compositions conferring high stability.

Purpose of the Study:

• To develop a deep learning model capable of designing highly thermostable proteins.
• To overcome the limitations of existing models in capturing the amino acid composition of hyperthermophilic proteins.
• To improve the stability of existing protein structures for practical applications.

Main Methods:

• Collected and predicted protein structures from hyperthermophiles.
• Retrained the ProteinMPNN network on hyperthermophile data to create HyperMPNN.
• Applied HyperMPNN to design protein nanoparticles and assessed their thermal stability.

Main Results:

• HyperMPNN successfully recovers the unique amino acid composition of hyperthermophilic proteins.
• The retrained network is applicable to proteins from non-hyperthermophiles.
• A protein nanoparticle designed with HyperMPNN maintained stability at 95°C, a significant increase from its original melting temperature of 65°C.

Conclusions:

• A novel self-supervised learning approach using hyperthermophile data enables the design of highly thermostable proteins.
• HyperMPNN represents a significant advancement in protein design for enhanced stability.
• This method has broad implications for developing robust protein-based therapeutics and biotechnologies.