The genomes are so five minutes ago. Personal medicine is about phenomena now.
Okay, it's an exaggeration. However, many genetic disorders lead to distinct facial phenotypes (Down syndrome is probably the best known example). Many of these disorders are quite rare and are therefore not easily recognized by clinics. This lack of familiarity can cause patients with the disease (and their parents) to endure a long and traumatic diagnostic odyssey before calculating what belongs to them. While they may be unusual, these rare diseases are generally not so rare: they affect eight percent of the population.
FDNA is a genomics / AI company that aims to "capture, structure and analyze complex human physiological data to produce viable genomic insights." They have made a facial image analysis framework, called DeepGestalt, which can diagnose genetic conditions based on on facial images with higher precision than doctors can. The results are published in Natural Medicine .
To train his algorithm, the company relied on a dataset of 500,000 facial images of 1
They then tested it by seeing how well it could identify faces in people with a certain genetic disorder when mixed in with faces of people with several other diseases – a situation that a clinician or genetic counselor could find very profitable in They did two tests of this type, one with Cornelia de Lange syndrome and the other with Angelman syndrome. Both are developmental disorders with cognitive and motor impairments. In both cases, DeepGestalt achieved more than 90 percent better precision than experts who were close to 70 or 75 percent accurate.
Another test was investigated if DeepGestalt could distinguish between a small pool of people with the same disease but different genotypes by showing that there are images of people with Noonan syndrome, which have a variable effect depending on which of five different genes are mutated . It only achieved 64 percent accuracy this time, but it is better than the 20 percent predicted by chance. Especially since "two dysmorphologists concluded that single phenotype was insufficient to predict the genotype."
The last test was to diagnose hundreds of images of faces spanning over 216 different disorders. That was 90 percent correct.
The algorithm works by pruning the face in several regions, determining how much each region corresponds to each syndrome and then aggregating the regions to see which syndrome is best suited. Therefore, Gestalt. But the authors state that "DeepGestalt, like many artificial intelligence systems, cannot explicitly explain its predictions and does not provide information on which facial features that drove the classification."
It's a black box; It can surpass experts in making a genetic diagnosis based on phenotype, but it cannot teach them how to do what it does.
Natural Medicine 2019. DOI: 10.1038 / s41591-018-0279 -0 (To DOIs).