Designing synthetic regulatory elements using the generative AI framework DNA-Diffusion.

Area of Science:

• Genomics
• Synthetic Biology
• Artificial Intelligence

Background:

• Precise control of gene expression via regulatory elements is a key challenge.
• Existing methods struggle to balance functional activity, cell-type specificity, and sequence diversity.

Purpose of the Study:

• Introduce DNA-Diffusion, a generative AI framework for designing novel regulatory elements.
• Achieve cell-type-specific gene expression control with compact synthetic elements.

Main Methods:

• Developed DNA-Diffusion, an AI framework trained on DNA accessibility data.
• Generated 200-base-pair synthetic elements mimicking endogenous transcription factor binding.
• Validated elements using STARR-seq across three cell lines and EXTRA-seq for endogenous gene modulation.

Main Results:

• DNA-Diffusion generated compact elements with enhanced cell-type specificity.
• Synthetic elements recapitulated endogenous transcription factor binding grammar.
• Successfully reactivated the AXIN2 gene in its native context, demonstrating functional modulation.

Conclusions:

• DNA-Diffusion outperforms existing computational methods in designing regulatory elements.
• This framework is crucial for engineering precise gene regulation in gene therapies.
• Enables advancements in understanding gene regulation and developing novel therapeutic strategies.