AI Decodes DNA 'Initiator' Sequence in 60% of Human Genes

3 min read
Source: Science Daily
AI Decodes DNA 'Initiator' Sequence in 60% of Human Genes
Photo: Science Daily
TL;DR

Researchers at UC San Diego used AI to decode the 'initiator,' a DNA sequence that acts as a switch for gene activation. By analyzing 500,000 variants, they found this element exists in roughly 60% of human genes, offering new tools to predict how mutations affect disease and gene expression.

Key points

  • UC San Diego researchers led by James T. Kadonaga and Torrey Rhyne-Carrigg decoded the DNA signature of the 'initiator,' a regulatory element that marks where gene transcription begins.
  • The team analyzed approximately 500,000 DNA sequence variants using high-throughput sequencing to train a machine learning model.
  • The AI model successfully predicted the presence or absence of the initiator in human genes, identifying it in about 60% of the genome.
  • This breakthrough allows scientists to better predict how mutations in regulatory regions may alter gene activity, potentially contributing to diseases like cancer.
  • The findings support the development of synthetic promoters for biotechnology and represent a step toward decoding the entire human gene expression code.

Background

This study builds on recent efforts to understand non-coding DNA and genome architecture. In August 2026, researchers mapped non-B DNA motifs across human and ape genomes, highlighting how DNA structure influences transcription. Earlier in 2025, studies on 'junk DNA' revealed switches linked to Alzheimer’s, while 2024 research identified mechanisms regulating disease-causing mutations. These findings collectively emphasize the growing role of AI in deciphering complex genetic instructions.

How outlets are covering it

ScienceDaily and Mirage News report the findings nearly identically, citing the same UC San Diego press release and emphasizing the 60% prevalence of the initiator. Bioengineer.org provides a more detailed technical perspective, explaining that the initiator does not follow a single universal sequence but appears in multiple forms. It highlights how the AI model captures functional relationships rather than just simple motifs, offering a more nuanced view of how regulatory DNA works. All sources agree on the potential for predicting mutation effects and designing synthetic promoters, but Bioengineer.org uniquely details the limitations of conventional motif searches compared to machine learning approaches.

Why it matters

Decoding the initiator helps explain why certain mutations cause disease without changing protein structures, focusing instead on when and where genes are activated. This knowledge could lead to better predictions of genetic risks and the creation of precise synthetic DNA switches for medical therapies, advancing personalized medicine and biotechnology.

What to watch

Researchers plan to expand AI models to cover the entire gene expression code, including enhancers and chromatin structure. The goal is to predict gene activity across different cell types and individuals, potentially leading to more accurate diagnoses and targeted treatments for disorders caused by regulatory mutations.

Share this article

Want the full story? Read the original reporting

Read on Science Daily