What will happen to our bodies if a letter changes in a person's DNA. Google DeepMind has analyzed a.. - 매일경제

It takes billions to search for disease variants

Unknown domain analysis, which accounts for 98% of rare diseases

Finding causes that were missed in research on rare diseases

What will happen to our bodies if a letter changes in a person's DNA. Google DeepMind has analyzed about 9 billion "one letter changes" that can occur in human DNA in advance with artificial intelligence (AI) and made it a huge map. It is expected that research time, which had to analyze numerous candidates one by one to find genetic mutations related to disease, will be greatly reduced.

Google DeepMind released "Alphagenome Atlas" on the 8th (local time), which predicted how about 9 billion possible one-letter changes in human DNA will affect our bodies. It is a kind of 'DNA mutation map'.

DNA stores genetic information by extending four types of bases: A, T, G, and C. Among them, the change of one letter to another is called a 'single base mutation'. For example, G enters where A should be. Most have little effect, but some may change the way genes work or be related to disease.

The problem is that there are about 9 billion possible changes in human DNA. It was virtually impossible to identify 9 billion mutations one by one in the laboratory to determine which mutations were important. Google DeepMind solved this with AI. Using the AI "alpha genome" released last year, we calculated in advance how each of the 9 billion mutations is likely to affect. Previously, researchers had to select variants they were curious about and analyze them with alpha genomes, but now they can search for the variant they want on the website.

"This is the first time any researcher in the world has access to a comprehensive map of the human genome and its mutations by opening a browser," said Pushmit Kohli, vice president of science at Google DeepMind.

Of particular interest to the alpha genome atlas is the region that accounts for 98% of human DNA. Only about 2% of DNA is directly used to make proteins. The remaining 98% do not make proteins themselves, but control when the gene will turn on and off and how active it will work.

This 98% region was difficult to interpret, although many disease-related variations were present. This is because no matter what mutation is found, it is not easy to figure out what it actually does in our body. Alpha Genome Atlas predicts which genes will affect when a letter in the region changes, and which cells and tissues will change.

Researchers have also created an "AVI score" to make it easier for them to select important mutations out of 9 billion. We scored how much each DNA mutation is likely to affect our bodies. Instead of looking at all of the billions of variations from the beginning, we can investigate the variations with high scores.

Ziga Absek, head of Google DeepMind Genomics, likened it to "finding a needle in a haystack." Researchers who do not know the cause of the disease can use AI to first select suspected candidates among numerous mutations, he explained.

In fact, rare disease studies have also shown promise. Researchers at the Broad Institute applied the AVI score to cases of rare diseases that have not been found in previous studies. As a result, we found a mutation that affects the DNM1 gene, which is closely related to epilepsy. AI predicted that this mutation could cause errors in the process of processing genetic information, resulting in abnormal proteins. This effect was also confirmed in later experiments.

It was also used in large-scale studies. Researchers at the University of Exeter in the UK analyzed the genome of more than 54,000 British Biobank participants using the Alpha Genome Atlas. As a result, we found a connection between DNA mutations and body characteristics, which were difficult to find in the existing method due to being buried in vast amounts of data. In the body mass index (BMI) study, the top 1% of mutations predicted by AI to have a large impact were analyzed intensively to find 19 possible genome regions.

However, AI does not conclude that "this mutation is the cause of the disease." Diseases are often caused by multiple genes and living environments working together. Alpha Genome Atlas is more of a tool to first select potential candidates among numerous mutations. When AI narrows down candidates, researchers must confirm them through real-world experiments.

Alpha Genome Atlas was released for free for non-commercial research. Google DeepMind said, "This technology has not yet been verified or approved for clinical use and cannot replace diagnosis or treatment by medical staff." [Silicon Valley correspondent Wonho-seop]