Decoding the "Code of Life"
Paving New Paths for Disease Diagnosis and Drug Discovery
Google DeepMind, which brought innovation to the scientific community last year with AlphaFold, an artificial intelligence (AI) for protein structure prediction that won the Nobel Prize in Chemistry, has taken on another major challenge: decoding the genome. There are growing expectations that this could have a profound impact on disease diagnosis, treatment methods, and new drug development.
On June 26 (local time), Google DeepMind introduced AlphaGenome, an AI capable of unlocking the secrets of DNA, in a published paper and released it as a research API. AlphaGenome can be downloaded and used by anyone for non-commercial purposes.
AlphaGenome can predict how small changes in DNA affect various molecular processes, such as increases or decreases in gene activity. It can process up to one million DNA characters and predict thousands of molecular traits related to gene regulation in just one second.
Just as AlphaFold revealed the 3D structure of proteins and transformed the landscape of life science research, AlphaGenome is expected to become the key to unlocking the secrets hidden in 99% of the genome, ushering in a new era in genomics.
The genome is the "instruction manual of the cell" that guides nearly everything about a living organism, including its appearance, functions, and growth. DNA contains human genetic information. Although the Human Genome Project mapped the human genome in 2003, understanding how the variations in the 3.1 billion DNA sequences operate at the molecular level has long remained a mystery. AlphaGenome was developed to solve this mystery.
DeepMind focused on the "regulatory genome," which plays a role in turning genes on and off. Mutations in the regulatory genome can be the cause of numerous diseases, including cancer, diabetes, heart disease, and autoimmune disorders. However, the language of the regulatory genome is extremely complex and vast, making it one of the greatest challenges in modern genetics to understand its function.
AlphaGenome is a large language model (LLM) that has learned the "grammar" of DNA. Just as ChatGPT uses the "transformer" architecture to learn grammar and context from human language data, AlphaGenome has mastered the "grammar" of gene regulation by training on vast amounts of DNA sequence data, according to the company.
Google DeepMind reported that AlphaGenome outperformed all existing prediction models and newly identified hundreds of thousands of previously unknown potential regulatory elements.
Pushmeet Kohli, Vice President of Research at Google DeepMind, said, "For the first time, we have created a single model that integrates the various challenges involved in understanding the genome."
Google DeepMind added that it plans to disclose all details of AlphaGenome and is also exploring ways for biotechnology companies to use it for commercial purposes in the future.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
