Controlling gene expression with deep generative design of regulatory DNA.

Area of Science:

• Synthetic biology
• Genomics
• Computational biology

Background:

• Designing synthetic regulatory DNA is crucial for biotechnology and medicine.
• Current methods using random mutagenesis are limited in scope and control.
• Efficiently controlling gene expression requires novel design strategies.

Purpose of the Study:

• To prototype a deep learning strategy for de novo synthetic regulatory DNA design.
• To generate gene-specific regulatory DNA with prespecified mRNA levels.
• To demonstrate the capability of deep generative design in controlling gene expression.

Main Methods:

• Utilized generative adversarial networks (GAN) trained on genomic and transcriptomic data.
• Developed ExpressionGAN to learn the regulatory sequence-expression landscape.
• Generated synthetic DNA sequences targeting specific mRNA levels across entire gene structures.

Main Results:

• ExpressionGAN successfully generated synthetic regulatory DNA with high sequence divergence from natural DNA.
• In vivo validation showed 57% of highly-expressed synthetic sequences exceeded natural controls.
• The method demonstrated control over gene expression across coding and non-coding regions.

Conclusions:

• Deep generative design offers a powerful approach to expand control over gene expression.
• ExpressionGAN enables precise, gene-specific design of synthetic regulatory DNA.
• This technology has broad applicability in biotechnology, medicine, and synthetic biology.