Sequence-based generative AI design of versatile tryptophan synthases

Area of Science:

• Biochemistry
• Protein Engineering
• Computational Biology

Background:

• Enzyme discovery and optimization are crucial for sustainable chemistry but face bottlenecks in identifying suitable starting points.
• Designing diverse and functional enzyme libraries remains a significant challenge in biocatalysis.

Purpose of the Study:

• To utilize a protein language model (GenSLM) for generating novel beta-subunit of tryptophan synthase (TrpB) enzymes.
• To assess the stability, catalytic activity, and substrate promiscuity of these generated TrpB enzymes.
• To demonstrate the potential of generative models in biocatalyst discovery and engineering.

Main Methods:

• Application of the GenSLM protein language model to design novel TrpB enzyme sequences.
• Expression and characterization of generated TrpB enzymes in Escherichia coli.
• Evaluation of enzyme stability, catalytic activity, and substrate promiscuity on native and non-native substrates.

Main Results:

• Successfully generated novel TrpB enzymes that express functional proteins in E. coli.
• Many generated TrpBs exhibited high stability and catalytic activity.
• Significant substrate promiscuity was observed, with some variants outperforming natural TrpBs and even laboratory-evolved enzymes.
• Comparison with natural homologs confirmed the acquisition of non-natural properties by generated enzymes.

Conclusions:

• Generative protein language models can create enzymes that retain natural functions while acquiring novel properties.
• These models represent powerful tools for accelerating biocatalyst discovery and engineering.
• The generated TrpBs demonstrate enhanced versatility, opening new avenues for enzymatic applications.