DNA serves as the fundamental blueprint for life, shaping everything from our physical traits to our long-term health. While scientists have long understood that genes—the genetic “words” that encode proteins—play a primary role in health and disease, these genes represent a tiny fraction of our genetic makeup . Surprisingly, more than 98 percent of the human genome consists of DNA that does not build proteins. Consequently, this vast region was once dismissed by researchers as “junk DNA”. Despite its importance, understanding exactly how noncoding DNA shapes gene expression has remained a significant mystery. To solve this puzzle, the AI lab Google DeepMind has developed a groundbreaking model called AlphaGenome .
This tool predicts the function of long stretches of noncoding DNA, offering insights into how these sequences influence human health. By analysing sequences of up to one million DNA letters (base pairs), the model predicts how mutations within those stretches affect the way genes express themselves.
Today, however, the scientific community recognises this “molecular dark matter” as a critical regulator of gene activity. Specifically, it determines whether genes are turned on or off, a process that is vital for keeping us healthy or, in some cases, causing chronic diseases.
AlphaGenome: A Window into the Dark Matter
The researchers described the model’s capabilities in a study published in the journal Nature. Furthermore, DeepMind has made a version of the tool freely available to the global research community. This accessibility will likely help scientists narrow down theories regarding how specific DNA changes impact gene function. In turn, this foundational knowledge could empower researchers to craft more effective treatments for complex genetic diseases.
Decoding the ‘Semantics’ of Life
Pushmeet Kohli, DeepMind’s Vice President for Science, explains that the company has long aimed to understand the “semantics” of the human genome. He compares the genome to a massive book containing three billion characters . If something goes wrong within those characters, AlphaGenome acts as a predictive tool to determine the consequences of those changes . Effectively, the AI can answer the question: “If you change these words, what would be the effect?”.
The Legacy of AlphaFold and AlphaMissense
AlphaGenome does not exist in a vacuum; rather, it is the latest in a series of highly successful AI models from DeepMind . It serves as a successor to AlphaFold, the revolutionary model that predicts the structure of nearly every known protein from its amino acid sequence . Notably, the researchers behind AlphaFold were awarded the Nobel Prize in Chemistry in 2024 for solving this central biological problem.
Additionally, AlphaGenome builds upon the foundations of AlphaMissense, a tool released in 2023. While AlphaMissense focuses on predicting how mutations in protein-coding regions affect gene function, AlphaGenome expands this capability to the much larger noncoding regions of our DNA . By combining information from various datasets focused on gene expression, AlphaGenome provides a holistic view of how our genetic instructions are executed.
Q&A: Understanding the Breakthrough
Why is noncoding DNA no longer considered “junk”?
Previously, scientists believed that DNA which didn’t create proteins had no purpose . However, recent research has proven that this DNA is crucial for controlling gene activity, acting as a control panel for our health .
How does AlphaGenome differ from AlphaFold?
While AlphaFold predicts the 3D structure of proteins, AlphaGenome focuses on the “instructions” found in noncoding DNA that tell the body when and how to build those proteins .
How can this AI help patients with genetic diseases?
By predicting how specific mutations affect gene expression, AlphaGenome allows scientists to identify the root causes of diseases more quickly . This leads to the development of targeted therapies and better medical outcomes.
Frequently Asked Questions (FAQ)
1. What is the primary function of AlphaGenome? AlphaGenome is an AI model designed to predict how changes or mutations in noncoding DNA affect the expression of genes .
2. How much of our DNA is noncoding? More than 98 percent of the human genome consists of noncoding DNA .
3. Is AlphaGenome available to the public? DeepMind has made a version of the AlphaGenome tool freely available to other researchers to accelerate scientific discovery .
4. What is a “base pair” in the context of this model? Base pairs are the “letters” that make up the DNA sequence. AlphaGenome can process and analyse sequences of up to one million of these letters at a time .
5. Who published the research on AlphaGenome? The model and its findings were described in the scientific journal Nature by researchers from Google DeepMind .