“These findings are another step in realizing the impact of AI on the natural sciences,” said Pushmeet Kohli, vice president of research at Google DeepMind.
The tool focuses on so-called “missense” mutations, where one letter of the genetic code is affected.
The entire genome of a normal human contains 9,000 such mutations; They can be harmless or cause diseases like cystic fibrosis or cancer or harm brain development.
To date, four million of these mutations have been identified in humans, but only two percent of them have been classified as either disease-causing or benign.
In total, there are 71 million such possible mutations. A Google DeepMind tool called Alphamisense reviewed these mutations and was able to predict 89 percent of them with 90 percent accuracy.
Each mutation was assigned a score, indicating the risk of causing disease (otherwise referred to as pathogenic).
The results: 57 percent were classified as probably benign and 32 percent as probably pathogenic — the remainder indeterminate.
The database was made public and available to scientists, and an accompanying study was published in the journal Science on Tuesday.
AlphaMissense exhibits “superior performance” over previously available tools, experts Joseph Marsh and Sarah Teichman wrote in an article published in Science.
“We must emphasize that the predictions were never trained or intended to be used only for clinical diagnosis,” said Jun Cheng of Google DeepMind.
“However, we think our prediction could potentially be useful in increasing rare disease diagnosis rates and helping to discover new disease-causing genes,” Cheng added.
Indirectly, this could lead to the development of new treatments, the researchers said.
The tool was trained on the DNA of humans and closely-related primates, enabling it to identify which genetic mutations are widespread.
Cheng said the training allowed the tool to “input millions of protein sequences and learn what a regular protein sequence looks like.”
It can then detect mutations and their potential for harm.
Cheng compared the process to learning a language.
“If we change a word from an English sentence, a person who knows English can immediately see whether changing the word will change the meaning of the sentence.”
la/arp/st