PHA synthase variant design using a conditional variational autoencoder

Area of Science:

• Biochemistry
• Enzyme Engineering
• Synthetic Biology

Background:

• Polyhydroxyalkanoate (PHA) synthases are dimeric enzymes crucial for synthesizing PHAs, biopolymers with diverse properties.
• Natural PHA synthases exhibit limited specificity, restricting the range of R-hydroxyacids they can polymerize.
• Tailoring PHA properties requires enzymes with broader or altered substrate specificities.

Purpose of the Study:

• To design novel PHA synthase enzymes with potentially enhanced or altered specificities using artificial intelligence.
• To validate the activity and functionality of computationally designed PHA synthases in a biological system.

Main Methods:

• A conditional variational autoencoder was employed to generate approximately 10,000 novel PHA synthase sequences.
• Sequences were selected based on conserved catalytic residues, dimer interface integrity, and structural features.
• In vivo validation of 16 selected enzyme variants was performed in yeast (Saccharomyces cerevisiae).

Main Results:

• Two novel PHA synthase variants, with 87 and 98 amino acid substitutions, were successfully designed and synthesized.
• Both active variants demonstrated the ability to produce poly(hydroxybutyrate) (PHB) when expressed in yeast.
• The study confirmed the efficacy of AI in creating functional variants of complex dimeric enzymes.

Conclusions:

• AI-driven protein design is a powerful approach for engineering complex enzymes like PHA synthases.
• Novel PHA synthases with significant modifications can be generated and shown to be active in vivo.
• This methodology opens avenues for creating tailored biopolymers with desired material properties.