AI Surpassed Evolution In Synthesizing Essential DNA Elements

Credit: Pixabay

In a new study, scientists demonstrated that generative AI models are capable of designing regulatory regions of DNA—the fragments responsible for gene expression. This represents a new step: instead of searching for suitable regulators in nature, the researchers began creating them from scratch for specific tasks.

The work is based on the DNA-Diffusion method, an algorithm that borrows ideas from diffusion models widely used in image and text generation. The model was trained on massive genomic datasets and learned to predict which DNA sequences would effectively turn on or enhance genes in specific cell types. As a result, the AI ​​was able to generate thousands of synthetic regulators, many of which proved to be functionally superior to their natural counterparts.

The work is based on the DNA-Diffusion method, an algorithm that borrows ideas from diffusion models widely used in image and text generation. The model was trained on massive genomic datasets and learned to predict which DNA sequences would effectively turn on or enhance genes in specific cell types. As a result, the AI ​​was able to generate thousands of synthetic regulators, many of which proved to be functionally superior to their natural counterparts.