Protein engineering using variational free energy approximation.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Protein engineering traditionally relies on Directed Evolution (DE), a time-consuming method.
• Generative deep learning models can design novel protein variants but often lack thermodynamic stability, yielding non-functional proteins.

Purpose of the Study:

• To introduce a deep learning model, PREVENT, for generating thermodynamically stable and functional protein variants.
• To evaluate the efficacy of PREVENT in designing variants of E. coli N-acetyl-L-glutamate kinase (EcNAGK).

Main Methods:

• Developed PRotein Engineering by Variational frEe eNergy approximaTion (PREVENT), a generative deep learning model.
• Trained PREVENT on protein sequence and structural datasets to learn thermodynamic landscapes.
• Applied PREVENT to design variants of the EcNAGK enzyme.

Main Results:

• PREVENT successfully generated 40 variants of EcNAGK.
• 85% of the designed variants were functional.
• 55% of functional variants exhibited growth rates comparable to the wildtype enzyme, even with up to 9 mutations.

Conclusions:

• PREVENT offers a significant acceleration in protein engineering compared to traditional methods.
• The model's ability to generate stable and functional proteins demonstrates a promising new direction for protein design.