AI model decodes DNA initiator sequence present in 60% of human genes

Scientists at the University of California San Diego have used artificial intelligence to decode the DNA sequence pattern of the initiator, a key segment where gene transcription begins. The study, led by graduate student Torrey Rhyne-Carrigg in Professor James T. Kadonaga's laboratory, was published in Genes & Development.

The team measured the gene expression activity of roughly 500,000 different initiator variants using high-throughput DNA sequencing. They then trained a machine learning model on this dataset to recognize the initiator's signature DNA pattern.

Applying the model to the human genome revealed that the initiator sequence is present in approximately 60% of human genes. According to Kadonaga, the AI models provide the first strong predictions of the presence or absence of the initiator in human genes.

The decoded pattern allows researchers to predict how DNA mutations within the initiator region may disrupt gene activation and contribute to disorders such as cancer. The data and models also offer a foundation for designing synthetic promoters with customized on-off functions.

Kadonaga described the work as a step toward combining laboratory experiments and AI to decipher the broader gene expression code embedded in human DNA. He noted that a complete AI model of this code could eventually predict the activity of gene variants across different individuals.

The study represents a specific advance in understanding core promoter architecture. While the initiator is only one component of gene regulation, its identification across a majority of human genes provides a new tool for functional genomics research.