Scientists at Google’s artificial intelligence arm DeepMind have developed a breakthrough tool that can predict whether genetic mutations are likely to cause harm – helping research into rare diseases.
A human on average has 9,000 genetic mutations throughout their genome. While most are harmless, some can cause cancer, cystic fibrosis or damage brain development.
Up until now, only four million mutations have been observed in humans by scientists, out of a possible 71 million such mutations. Additionally, only two per cent have been classified as disease-causing or benign.
How does the AI tool work?
The Google AI tool, named, AlphaMissense, makes predictions about so-called missense mutations, where a single letter is misspelt in the genetic (DNA) code. Upon reviewing the mutation, the AI tool was able to predict 89 per cent of them, with 90 per cent accuracy.
When the researchers set the programme’s precision to 90 per cent, it predicted 57 per cent of missense mutations were probably harmless and 32 per cent were probably harmful while remaining uncertain about the rest.
To teach the AI system, scientists fed DNA data from humans and closely associated primates which enabled it to learn about the missense mutations that are relatively common and, therefore benign and others which are rare and potentially dangerous.
While the scientists taught the system, it familiarised itself by studying millions of protein sequences and learning what a “healthy” protein looks like.
Upon being fed a mutation, the AI tool generates a score which is assigned to each mutation, indicating the risk of it causing disease (otherwise referred to as pathogenic).
“This is very similar to human language. If we substitute a word in an English sentence, a person familiar with English can immediately see whether the word substitution will change the meaning of the sentence or not,” said Jun Cheng of Google DeepMind.
The database compiled by AlphaMissense has now been made public and available to scientists, accompanying a study that was published in the journal Science.
Scientists are of the view that the breakthrough could lead to the development of new treatments and assessing the risk of a deadly mutation before it reaches a point of no return.
“We should emphasise that the predictions were never really trained or never really intended to be used for clinical diagnosis alone,” said Jun.
“However, we do think that our predictions can potentially be helpful to increase the diagnosed rate of rare diseases, and also potentially to help us find new disease-causing genes.”
While the AI tool is able to predict if a genetic mutation is harmful, the onus remains on the human doctors to understand the reason and ultimately use the information to operate on the patient.
(With inputs from agencies)