Two scientists from Google DeepMind shared the 2024 Nobel Prize in chemistry for developing AlphaFold2, an artificial intelligence system that predicts the three-dimensional structures of proteins. The breakthrough solved a problem that had challenged researchers for decades and quickly became a staple in laboratories worldwide.
“Everyone’s using AlphaFold,” said Alex Palazzo, a geneticist at the University of Toronto.
Building on that success, Google DeepMind has introduced a new AI system aimed at deciphering DNA. The tool, called AlphaGenome, was unveiled last week in the journal Nature. Researchers say it has been trained on vast amounts of molecular and genetic data, enabling it to predict how mutations may influence the activity of thousands of genes.
One of AlphaGenome’s key capabilities is forecasting whether a mutation will switch a gene off or activate it at the wrong time, a factor central to understanding cancer and other diseases. Peter Koo, a computational biologist at Cold Spring Harbor Laboratory who was not involved in the research, described the system as “an engineering marvel.”
Still, outside experts urged caution. “This is not AlphaFold, and it’s not going to win the Nobel Prize,” said Mark Gerstein, a computational biologist at Yale University. He added that while the tool will likely become part of many researchers’ toolkits, it represents an early step in a much larger scientific journey.
Understanding DNA has long posed formidable challenges. Genes are written in a four-letter code of bases, and cells read these sequences to produce proteins. Yet the genome is far more complex than once believed. Through a process called splicing, cells can rearrange genetic instructions to produce multiple proteins from a single gene. Errors in splicing are linked to numerous diseases.
Gene regulation presents another puzzle. Specialized molecules bind to DNA and fold it into loops, sometimes exposing genes for activation and other times hiding them. These control regions can sit thousands or even millions of bases away from the genes they influence.
DeepMind researchers began work on AlphaGenome in 2019, training earlier models such as Enformer before expanding to the current system. AlphaGenome now makes predictions across 11 biological processes and has matched or exceeded other AI tools in performance tests, according to the study.
In one example, the system correctly predicted how mutations far from the TAL1 gene could trigger its constant activation, a change linked to leukemia. “It was quite mind-blowing,” said Dr. Marc Mansour of University College London.
Yet scientists stress that predictions must be confirmed in the lab. “These prediction tools are still prediction tools,” Mansour said.
While AlphaGenome signals progress in applying artificial intelligence to genomics, researchers agree that translating such tools into clinical practice remains a distant goal.
