Google's AlphaGenome AI Model Decodes Human Genome's 'Dark Matter' for Disease Insights

A new artificial intelligence model from Google’s DeepMind could soon help scientists read the most cryptic parts of the human genome, the so-called ‘dark matter’ of DNA, potentially transforming our understanding of diseases, drug discovery, and how life itself is written. The model, named AlphaGenome, has been hailed as a breakthrough in genomics research. It can predict how even the tiniest changes in DNA, a single swapped letter in the genetic code, might alter the way genes behave, and in turn, how those changes could contribute to conditions such as cancer, obesity, dementia, and high blood pressure. Researchers say it marks a giant leap toward decoding the biological “operating system” that powers the human body. The human genome is an enormous book of life, written with just four letters, A, C, G, and T, across three billion positions. But only about 2 per cent of that text contains genes that directly code for proteins. The remaining 98 per cent, long dismissed as “junk DNA”, has proven to be far from useless. Scientists now call it the dark genome, a hidden network of sequences that control when, where, and how genes are switched on or off. It’s also where many mutations linked to complex diseases tend to lurk. Until now, decoding that 98 per cent has been one of biology’s hardest puzzles. That’s where DeepMind’s AlphaGenome steps in. Unlike chatbots that predict the next word in a sentence, AlphaGenome is what researchers call a “sequence-to-function model”. Instead of guessing text, it learns how changing one piece of DNA can affect the biological meaning, what proteins get made, which genes activate, and how those changes ripple through cells. DeepMind says the model can analyse one million letters of DNA code at a time, far beyond the capacity of traditional methods. It can identify genes, predict how mutations alter gene expression, and even forecast how cells splice genetic instructions to build hundreds of proteins from a single gene. “We see AlphaGenome as a tool for understanding what the functional elements in the genome do,” said Natasha Latysheva, a DeepMind research engineer, in an interview with the BBC. “We hope it will accelerate our fundamental understanding of the code of life.” AlphaGenome follows in the footsteps of DeepMind’s AlphaFold, the revolutionary system that won the 2024 Nobel Prize in Chemistry for predicting protein structures, a task scientists had struggled with for decades. AlphaFold reshaped biology by helping researchers design new proteins and study how malfunctions in folding can lead to disease. Now, DeepMind wants to achieve the same for DNA. The company’s new AI was trained on vast molecular datasets, including the results of thousands of gene activity experiments, allowing it to make predictions about DNA segments it’s never seen before. The model was first made available to researchers last year, and more than 3,000 scientists have already used it in projects ranging from studying obesity and diabetes to uncovering cancer-linked mutations. By predicting which genetic variants are truly driving disease, AlphaGenome could help scientists focus on the most promising targets for drug development. Experts believe the model’s ability to identify where mutations disrupt normal gene function could lead to breakthroughs in synthetic biology and gene therapies, enabling researchers to design entirely new DNA sequences with desired effects. Still, the technology isn’t flawless. DeepMind admits that AlphaGenome is less accurate when it comes to predicting long-range gene regulation, where one piece of DNA influences another thousands of letters away, or understanding gene activity across different tissues. The team plans to refine those capabilities in future iterations. Despite the limitations, scientists are calling AlphaGenome “a big leap” for genomics. “It adds another piece of the puzzle for discovering new drug targets and ultimately developing better treatments,” said Latysheva. If AlphaFold helped scientists understand the shape of life, AlphaGenome could help them decode its instructions, the invisible logic that governs how every cell functions, evolves, and sometimes fails. The race to read life’s most mysterious script just gained a powerful new interpreter. Alibaba will spend 3 billion yuan ($431 million) on incentives for its Qwen AI app during the Lunar New Year, outpacing Tencent and Baidu’s similar campaigns, as Chinese tech firms compete for users with digital red envelopes and promotions. Get the latest stories delivered straight to your inbox.