GeoEvoBuilder: A deep learning framework for efficient functional and thermostable protein design.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Protein engineering traditionally relies on iterative computational design and experimental validation, which is time-consuming.
• Existing deep learning methods have advanced protein design but struggle to directly generate sequences with desired properties like high activity and stability.

Purpose of the Study:

• To present GeoEvoBuilder, a novel deep learning framework for direct protein sequence design.
• To integrate structural and evolutionary constraints adaptively for improved protein engineering outcomes.

Main Methods:

• Developed GeoEvoBuilder, a deep learning framework integrating structural and evolutionary information.
• Applied GeoEvoBuilder to redesign green fluorescent protein, glutathione peroxidase 4 (GPX4), and dihydrofolate reductase (DHFR).
• Validated designed protein structures and functions through experimental methods, including crystal structure determination.

Main Results:

• GeoEvoBuilder accurately predicted functional sites and generated stable, active protein sequences.
• Redesigned DHFR showed a 20-fold increase in catalytic efficiency and 10 °C higher thermal stability.
• Designed proteins exhibited correct structures, with GPX4 variants offering insights into residue dynamic correlations and enzymatic activity regulation.

Conclusions:

• GeoEvoBuilder offers a transformative approach to protein engineering by enabling direct sequence design with significant improvements in activity and stability.
• The framework successfully redesigned multiple proteins, demonstrating its versatility and potential for exploring vast sequence spaces with substantial residue modifications.
• GeoEvoBuilder facilitates a deeper understanding of protein sequence-structure-function-evolution relationships and is publicly available for research.