UCSD Team Uses AI to Unveil Rare DNA Sequences with Custom Functions

Researchers at the University of California San Diego used machine learning to delve into the intricacies of gene activation, a critical process involved in growth, development, and disease. James T. Kadonaga, senior author of the paper published in Genes & Development, and his colleagues previously employed AI to identify a puzzling component associated with gene activation known as the downstream core promoter region (DPR). This enigmatic DNA sequence acts as a "gateway" code that regulates the operation of approximately one-third of human genes.

Building upon their previous breakthrough, the team used machine learning to identify “synthetic extreme” DNA sequences with specifically designed functions in gene activation. They did this by testing millions of different DNA sequences through machine learning by comparing the DPR gene activation element in humans versus fruit flies (Drosophila). By using AI, they were able to find rare, custom-tailored DPR sequences that are active in humans but not fruit flies and vice versa.

Kadonaga envisions numerous practical applications for this strategy. It could be employed to identify synthetic extreme DNA sequences tailored for specific purposes, such as activating a gene in one tissue but not another or testing the effects of different drugs. Kadonaga further suggests that these rare DNA sequences, if they exist, might be found using AI, even if they occur at a frequency as low as one in a million.

To train their machine learning models, the researchers employed a method called support vector regression and used data from real-world laboratory experiments involving 200,000 established DNA sequences. These sequences served as examples for the AI system, which was then tasked with analyzing an extensive dataset of 50 million test DNA sequences for humans and fruit flies. The AI models successfully identified unique DNA sequences specific to each species.

Importantly, the predicted functions of these extreme sequences were verified in Kadonaga's laboratory using conventional wet lab testing methods. According to Kadonaga, it would be practically impossible to conduct the equivalent of 100 million wet lab experiments due to the extensive time required for each experiment.

The successful identification of rare DNA sequences through machine learning serves as a testament to the power of AI in biology. Kadonaga believes that this approach has broader implications for designing customized DNA elements in gene activation, with potential applications in biotechnology and biomedical research.