UC San Diego researchers used machine learning trained on half a million experimental DNA variants to crack the sequence code of the “initiator” — a core genetic element that helps trigger gene activation — a finding that could help predict how mutations affecting it contribute to disease.
Precise activation of the tens of thousands of genes in the human genome, at the right time and in the right cell type, underlies healthy development; when this activation process goes wrong, the consequences range from developmental disorders to cancer. A specialised DNA element called the “initiator” — which marks where a gene’s genetic information begins to be converted into a functional product — is known to play a central role in this process, but until now its precise sequence signature had not been comprehensively decoded.
Researchers in the laboratory of UC San Diego professor James T. Kadonaga, led by graduate student Torrey Rhyne-Carrigg, used high-throughput DNA sequencing to measure the gene-expression activity of roughly 500,000 different variants of the initiator sequence. They then trained a machine-learning model on this dataset to identify the characteristic DNA pattern the initiator follows. “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. Once the model had learned this signature, the team searched the human genome for matching sequences and found the initiator present in approximately 60% of human genes — a substantially larger share than had previously been established with confidence.
Why the method matters as much as the result
The scale of the underlying experiment — half a million tested variants, analysed with a machine-learning model rather than conventional statistical methods — reflects a broader shift in molecular biology toward using AI to find patterns in large combinatorial datasets that would be impractical to analyse by hand. The resulting model does more than describe the initiator retrospectively: because it can predict the presence or absence of a functional initiator from sequence alone, it gives researchers a new practical tool for anticipating how a mutation in a previously uncharacterised gene region might disrupt gene activation, without needing to test that specific gene experimentally.
Why it matters
Gene misregulation — genes switched on or off incorrectly — underlies a wide range of human diseases, including many cancers, and identifying the sequence code an activation element follows is a prerequisite for predicting the functional consequences of mutations that fall within it. With roughly 60% of human genes now confirmed to carry an initiator, this is not a niche regulatory element but one of the most common gene-activation mechanisms in the human genome — meaning the predictive model has broad potential application across disease genetics. The research also has a stated synthetic-biology application: the same dataset and AI model could support the design of synthetic promoters, DNA sequences engineered to drive gene expression in biotechnology and gene-therapy applications, where a reliable, predictable on-switch is a basic design requirement.
Dr. Ismail S Penugonda
Key facts
- The “initiator” is a DNA sequence element that helps trigger gene activation, present in roughly 60% of human genes according to the new AI model
- Model trained on high-throughput sequencing data from ~500,000 tested initiator variants
- Published in Genes & Development; led by Torrey Rhyne-Carrigg in the laboratory of Professor James T. Kadonaga, UC San Diego
- Potential applications include predicting disease-relevant mutation effects and designing synthetic promoters for biotechnology
