Our bodies are in a constant state of activity on a scale that is almost unfathomable. Trillions of genetic switches are being turned on and off every day as our cells respond to the world around us and to the changing needs of our bodies. When the sun’s UV rays hit the skin, for example, cells activate genes involved in producing melanin, helping protect the body from further damage. When a virus enters a cell, genetic machinery responds by activating genes that help produce interferons and other proteins involved in the immune response.

The complexity of it all, trying to understand how specific regions of DNA help coordinate this process, has long challenged researchers. Now, scientists at the University of California San Diego have used AI to crack a key piece of DNA’s hidden “on switch,” bringing researchers closer to understanding how the human genome controls when genes are activated.

The breakthrough, announced Aug. 21 by UC San Diego, centers on a crucial stretch of DNA known as the “initiator,” a sequence that helps specify where the instructions encoded in genes begin the process of being transcribed into RNA, and ultimately used to produce the proteins and other molecules essential to life.

The ability to identify the initiator’s DNA sequence gives scientists a new way to search the human genome for mutations that could disrupt gene activity and contribute to diseases, including cancer, according to the UC San Diego researchers.

Researchers in the laboratory of UC San Diego molecular biology professor James T. Kadonaga first generated an enormous amount of experimental data, measuring the gene-expression activity of approximately 500,000 different versions of the initiator. They then used that information to train a machine-learning model capable of recognizing the DNA pattern associated with a functioning initiator, according to the study published in Genes & Development.

DNA is not simply a long string of genetic instructions. It is more like a complicated operating system, containing the instructions for making proteins and other molecules as well as the regulatory information that tells cells when, where and how strongly those instructions should be used. Getting that timing wrong can contribute to disease.

The initiator is part of the machinery at the beginning of that process. Scientists have known about the initiator for years, but the precise sequence features that define it across human DNA have been difficult to determine.

The machine-learning model was able to sift through the data and identify the sequence characteristics that distinguish an initiator.

The researchers, led by graduate student Torrey Rhyne-Carrigg, found that about 60% of human genes have an initiator sequence in their core promoters, giving scientists a much clearer picture of how this particular component of the gene-activation system is distributed throughout the genome, according to UC San Diego.

“These AI models were found to provide, for the first time, strong predictions of the presence or absence of the initiator in human genes, and were thus able to decode the DNA base sequence pattern of the initiator,” Kadonaga said in a statement released by the university.

AI has become such a powerful tool in biological research because it can help researchers make sense of enormous amounts of biological information that would be difficult, if not impossible, for humans to analyze on their own.

One of the most striking examples is AlphaFold, the AI system developed by Google DeepMind that transformed the long-standing problem of predicting the three-dimensional structures of proteins. Before AlphaFold, determining the structure of a single protein could take years and require enormous amounts of time and money. AlphaFold demonstrated in 2020 that AI could predict protein structures with remarkable accuracy, opening a new avenue for researchers trying to understand how proteins work and how they interact with other molecules, according to Google DeepMind.

In 2024, AlphaFold’s creators Demis Hassabis and John Jumper were awarded the Nobel Prize in Chemistry for their work on protein structure prediction.

The UC San Diego research is addressing a different problem, but the principle is similar. Biology is filled with patterns that contain information about how life works, and AI can help scientists find those patterns when the amount of data becomes too vast for traditional methods to handle efficiently.