Researchers at Google DeepMind claim to have built an AI model that can pinpoint which genetic mutations can cause disease. new research in diary science.
The new model, called AlphaMissense, is an adaptation of DeepMind’s breakthrough AlphaFold, which in 2020 finally solved the protein folding problem that has long puzzled the scientific community. AlphaMissense is “fine-tuned” for genetic differences between “humans and primates” and specifically trained to identify “missense” mutations, or genetic mutations that occur in a single letter of the DNA code, according to a new study. It is said that it has been done.
Some missense mutations are completely benign, but any human Around 9,000 Missense alleles in DNA — other alleles can cause serious disease. Sickle Cell Anemia, Cystic Fibrosis, and Cancer with DeepMind Pointed out in Tuesday’s blog post, all specifically from missense genes. However, despite the fact that missense mutations and other DNA abnormalities are the main drivers of disease, humans are only able to independently classify just 0.1% of missense genes as good or bad.
Until now, yes. According to new research from DeepMind, this new AI model was able to identify a staggering 71 million missense mutations, and from there classified 89% of these mutations as “likely benign or pathogenic.” We were able to predictably classify the situation as ‘very likely’. Tens of millions of these predictions are then spun up into a vast online database for doctors, genetic researchers, and other diagnostic experts, and Google says they can use this new resource to It is expected that it will be possible to detect and diagnose a variety of diseases, including diseases. Eventually, they would begin developing what they called a “life-saving treatment.”
“Today, we are publishing a catalog of ‘missense’ mutations, where researchers can learn more about what impact they may have,” DeepMind said on Tuesday. “By using AI predictions, researchers can get a preview of the results,” they wrote in a blog post, adding that “by using AI predictions, researchers can get a preview of the results.” Thousands of proteins can be analyzed at once, allowing them to prioritize resources and Help accelerate your research. ”
But while that’s all sound As great as it is, this news received mixed reactions from the scientific community.
People like Ewan Barney, deputy director of the European Institute for Molecular Biology, said to BBC Calling it a “major step forward,” AlphaMissense claimed that the model “will help clinical researchers decide where to focus their efforts to find areas that may cause disease.” But others, like Ben Lehner, senior group leader in human genetics at Britain’s Wellcome Sanger Institute, were more hesitant. tell guardian The black box aspect of technology concerns him.
“One of the concerns about the DeepMind model is that it’s very complex,” Lehner said. guardian. “Models like this may turn out to be more complex than the biology they are trying to predict,” he added, because doctors cannot predict. Really Understanding how models like AlphaMissense actually work and using their predictions to make diagnostic choices can be problematic.
“It’s humbling to know that we may never understand how these models actually work. Does this matter?” Lehner said. guardian. “This may not apply to some applications, but can physicians feel comfortable making decisions about patients they cannot understand or explain?”
However, that being said, Lehner said the DeepMind model “does a good job of predicting what’s broken” and that “knowing what’s broken is a good first step.” Still, “if you want to fix it, you also need to know how something broke,” he says.
Of course, AlphaMissense isn’t quite there yet. After all, genetics are endlessly complex. “We’re excited to announce that this is the first time we’ve seen this,” said Heidi Rehm, director of clinical laboratories at the Broad Institute at the Massachusetts Institute of Technology and Harvard University. Said MIT Technology ReviewComputer predictions are just one piece of evidence for doctors to make a diagnosis.
“Models are improving, but none of them are perfect. We still don’t know if they’re pathogenic,” Rehm continued, noting that Google is overstating the medical efficacy of new products. “I’m disappointed,” he reportedly said.
Therefore, there are mixed reviews. But even if DeepMind’s purported progress is not; very The venture has had a big run so far, but it still has plenty of potential to move forward. Only time will tell, but in the meantime, if you work in diagnosing genetic diseases, you might want to take AlphaMissense’s predictions with a grain of salt.
More about medical innovation: Biotech company says it has transplanted dopamine-producing cells into patients’ brains