Select Language:
Google introduced a new artificial intelligence tool on Wednesday designed to help unlock the secrets of the human genome. According to Google scientists, this technology could eventually lead to innovative treatments for genetic diseases. The deep learning model, named AlphaGenome, has been praised by external researchers as a significant breakthrough that allows scientists to analyze and even simulate the origins of complex genetic disorders.
Although the first complete human genome map was created in 2003, understanding how to interpret this vast amount of data—often called the “book of life”—remains a challenge, noted Pushmeet Kohli, Vice President of Research at Google DeepMind. “We have the sequence of three billion nucleotide pairs represented by the letters A, T, C, and G, which compose DNA,” he explained. However, grasping the ‘grammar’ of this genome—the way the information encoded within it influences biological functions—is the next important step, he added, citing a study published in Nature.
Only about 2% of our DNA comprises instructions for making proteins, the molecules responsible for constructing and maintaining the body. The remaining 98% was once dismissed as “junk DNA,” but scientists now believe this non-coding DNA acts like a conductor that orchestrates how genetic information functions across cells. These sequences also contain numerous variants linked to various diseases, and AlphaGenome aims to decipher these relationships.
The project is part of Google’s broader efforts in AI-based scientific research, which also includes AlphaFold, awarded the 2024 Nobel Prize in Chemistry. AlphaGenome’s model was trained on data from public projects measuring non-coding DNA across hundreds of human and mouse tissues and cell types. It can analyze extensive DNA sequences—up to a million nucleotides long—and predict how each pair influences cellular processes, such as gene activation, protein production, and RNA synthesis.
While similar models exist, they often analyze shorter sequences or offer less detailed predictions. Ziga Avsec, lead author of the study and a scientist at DeepMind, emphasized that longer sequences are essential to fully understand the regulatory environment of individual genes. The high-resolution capabilities of AlphaGenome enable scientists to examine how genetic variants—mutations—affect gene function by comparing mutated and normal sequences.
“AlphaGenome accelerates our understanding of the genome by pinpointing where functional elements are located and revealing their roles at a molecular level,” added co-author Natasha Latysheva. The tool has already been utilized by approximately 3,000 scientists worldwide and is available for non-commercial use, with Google encouraging researchers to contribute additional data to enhance its capabilities.
Industry experts who tested the system acknowledge that it performs well but cautioned that it’s not a cure-all. Ben Lehner from Cambridge University, who wasn’t involved in the development but has tested AlphaGenome, noted that accurately identifying genetic differences related to disease risk is crucial for developing more effective treatments. However, he also highlighted that the model is far from perfect and that the training data’s limitations currently constrain its accuracy.
Robert Goldstone, head of genomics at the UK’s Francis Crick Institute, pointed out that AlphaGenome isn’t a magic solution for every biological question because gene expression is heavily influenced by environmental factors that the model cannot account for. Nonetheless, he recognized the tool as a significant breakthrough that will assist scientists in studying and simulating the genetic roots of complex diseases.
ChatGPT
Perplexity
Gemini AI
Grok AI
