Generative design of synthetic gene circuits for functional and evolutionary properties

Area of Science:

• Synthetic Biology
• Computational Biology
• Machine Learning

Background:

• Designing reliable and predictable biotechnologies from genetic circuits remains challenging despite available tools.
• Machine learning and modeling aid in navigating genetic circuit design and predicting component functions.
• Generative algorithms are emerging for designing genetic circuits from sequence level, with recent applications in synthetic biology.

Purpose of the Study:

• To demonstrate that simple generative models, like conditional variational autoencoders (CVAE), can design novel genetic circuits matching complex dynamic functions.
• To explore the potential of generative models in synthetic biology for creating adaptable circuits.
• To investigate the simultaneous design of functional and evolutionary properties (phenotype and evotype).

Main Methods:

• Utilized in silico RNA simulation to create a dataset of RNA sequences.
• Employed RNA interaction predictors to convert sequences into genetic circuits.
• Estimated functional features, evolutionary stability, and interpreted model-learned features.
• Framed signal adaptation as a single design objective for model training.

Main Results:

• The conditional variational autoencoder (CVAE) model successfully generated diverse genetic circuits exhibiting signal adaptation.
• Model-generated circuits effectively matched target adaptation specifications, even with limited training data.
• The model's embedding space structures corresponded to known adaptation motifs and revealed design rules.
• Simultaneous prompting of functional and evolutionary properties was demonstrated, showing combined phenotype and evotype design.

Conclusions:

• Simple generative models are effective for designing complex genetic circuits in synthetic biology.
• The choice of data encoding significantly impacts the generation of functional genetic circuits.
• This work provides a proof-of-concept for designing both phenotype and evotype simultaneously in genetic circuits.