Some people’s faces — or even just a photo of them — hint at the genes they carry. And now, an algorithm can predict not only whether they carry a genetic mutation, but which genes were mutated.
The study, published Monday in Nature Medicine, is the latest from a Boston-based company called FDNA, one of a few organizations creating software that can help physicians diagnose genetic syndromes based just on a face — and may serve an important validation of the company’s technology, said Yaron Gurovich, the company’s chief technology officer.
“We went for this high-impact journal to prove beyond any doubt that this technology is good, it performs as we say, we can stand behind it, and now it opens a lot of doors to publish more,” he said.
The study itself is a collection of experiments testing how the results of algorithms — FDNA refers to them as DeepGestalt — stack up against clinicians’ diagnoses. In one of the experiments, DeepGestalt’s performance was better than random chance when picking which of five genetic mutations might be causing a condition called Noonan syndrome. It was correct 64 percent of the time, far more than the 20 percent success rate that would be expected from guesswork.
“This is new — we’ve never published something like this before,” Gurovich said.
Gurovich is quick to say that the tool isn’t specifically or only for Noonan syndrome. His team chose the condition because there are already published studies about how well humans can distinguish between the various faces associated with it. FDNA is already working on another paper Gurovich said will show that the tool can be used more broadly. It’s going through the peer review process, he said, but a preprint version is available.
One expert on Noonan syndrome, Dr. Bruce Gelb, the director of the Mindich Child Health and Development Institute and a professor at the Icahn School of Medicine at Mount Sinai, cautioned that being able to pick apart a person’s genotype based on facial features is not generally going to be useful for people with the condition.
Noonan syndrome comes with a variety of symptoms, including difficulty learning, facial appearance, short stature, and heart issues — including issues with valves or the muscles of the heart itself. A few have a very high risk of leukemia.
Some children with Noonan syndrome attend special education classes; others develop typically and can attend mainstream classes. Many can live independently when they’re adults. “It varies a lot,” Gelb said.
The genetic cause of Noonan syndrome can vary, too. Mutations in a few different genes can lead to the condition; some mutations cause more serious problems than others. All of the genes, however, are linked to one vital biochemical pathway. Gelb and his research group have discovered some of them.
For example, children whose RAS1 gene is mutated almost always get hypertrophic cardiomyopathy, a condition in which the muscles of the heart get thick, making it difficult for the heart to pump normally. Children with mutations in a gene called KRAS have some of the most severe forms of the syndrome and some of the worst neurological and heart outcomes.
There aren’t any drugs to treat Noonan syndrome, or many other developmental syndromes like it.
Understanding how a child with Noonan syndrome will develop can help health care providers figure out what medical problems they may face, Gelb said. But the algorithm isn’t likely to replace a genetic test, he said — which doctors can undertake easily if they notice something off in a particular region of a fetus’ neck.
“I don’t know why they undertook this, exactly,” he said of FDNA’s work. “It’s inconceivable to me that one wouldn’t send off the panel testing and figure out which one it actually is.” Even in low-resource countries — at least in those with a medical geneticist — such genetic testing is becoming more widely available, he said.
Gelb also pointed out that the paper only used a set of images of young children — a choice that may have set the algorithm up for success. “The facial features are most obvious in a toddler or young child, and it can kind of melt away in adolescence by the time they hit adulthood,” he said.
He acknowledged that the algorithm’s success rate, however, is “impressive” and could be especially useful for clinicians who don’t have hyper-specialized knowledge about a given genetic condition.
And using a tool like FDNA’s could show clinicians what genes they should ask labs to test, Gurovich suggested. “If you consider the phenotype properly, you are able to increase your odds of a diagnosis,” he said — something that he said humans can’t quite do.
“There are geneticists that have tried to do this. They couldn’t. We can.”