Dysmorphology: even sophisticated medical consumers are unfamiliar with the term. It refers to a method of medical diagnosis whereby serious underlying genetic conditions are detected through examining facial and bodily deformities. It was an art long essential in diagnosing a large cluster of rare and otherwise hard-to-detect conditions through phenotyping — a last resort, so to speak, of a patient desperate for an accurate diagnosis.
The discipline, despite its necessary place in the medical toolkit, had fallen into disuse as more and more sophisticated diagnostic methods focused on the patient’s genotype (as opposed to phenotype) proliferated through the late 20th and early 21st centuries. Moti Shniberg and Dekel Gelbman, two of the founders of FDNA, have changed all that. By developing and implementing a powerful facial recognition tool backed by a deep database of facial deformities associated with various severe, rare, and hard-to-diagnose syndromes — thus creating the app Face2Gene — they have not only revived phenotyping as a diagnostic art. They have made diagnosing a whole group of conditions easier, saving and improving lives in the process. Best of all? Their technology does its work in about the time it takes to order an Uber.
Shniberg came to phenotyping from the world of serial entrepreneurship. He’s launched products and companies in spheres as varied as machine vision and the art world, where he developed a revolutionary fund to help artists survive financially by “investing” their own works. Gelbman has a more traditional business background, doing high-level M&A deals, VC funding, and IPO work at Skadden Arps. But the potential and mission of FDNA drew them both in — as it did the third founding member, Dr. Lior Wolf, a topline computer scientist at Tel Aviv University with major background in deep learning. As Gelbman put it in in our exclusive interview with him and Shniberg, “The name is a bit confusing: rare diseases. Because while each particular disease in itself is extremely rare, when you aggregate them, we’re talking about more than 8,000 different syndromes that can affect up to one in ten people on the planet — a huge population. What we found is that the patient’s journey in these cases is really a long journey. It starts with a lot of people being affected with a wide range of diseases and a lack of awareness and good tools among clinicians to pick up the symptoms and correlate them with underlying genetic syndrome. It continues with molecular testing. Although it’s more affordable and more accessible, if you look at exome sequencing [the sequencing of the part of the human genome that codes for proteins], the diagnostic yield is only about 25 percent. It’s not a catch-all test as people assumed. And given the fact the fact that it takes seven years to diagnose a patient, there’s a scarcity of patients for drug companies to recruit for clinical trials and to identify treatments for these populations.”
Dekel’s theory of the market for this technology has been proven abundantly correct. As of press time, some 70 percent of the world’s geneticists use FDNA’s technology, and it can be found in 50 percent of genetics labs globally. In short, it is unquestionably the dominant product in the space at the moment. With good reason. The technology combines cutting-edge machine learning and visualization tools — something Shniberg knows a little bit about; he was the man who developed Facebook’s revolutionary tagging technique for photos — with a database that Gelbman calls “100 percent clinically relevant — we don’t have just normal individuals.” The basic process through which the tech works is simple. A physician presented with a complex of chronic symptoms and deformities will usually run through tests based on genotype. These, though accurate and powerful, result as Gelbman noted in a poor diagnostic yield for the population FDNA’s database focuses on. Facing inconclusive results some 75 percent of the time, the physicians would often turn in such cases to a specialist in dysmorphology. But now matter how skilled any single dysmorphologist might be, his knowledge was limited by experience. FDNA, in effect, allows physicians to submit their patients profiles to an AI-powered dysmorphologist with a vast and monumentally relevant store of “experience” — in this metaphor, the database. As Michael Hayden, the eminent geneticist and current head of research and development at Israeli pharma major Teva, put it, FDNA has “provided an evidence-based science to what was previously considered the art of dysmorphology.”
Though the company is now the top phenotyping platform in the world and seems poised for major success — they raised more than $30 million in a recent oversubscribed funding round — both Gelbman and Shniberg see the project as having a philanthropic aim as much as a commercial one. Shniberg expressed this quite clearly: “I believe that above what we are doing on rare disease is very important. We are talking about eight percent of the global population. It’s having a great impact on saving people’s lives.” Gelbman is equally enthusiastic about the social value of the technology: “The mission really resonated in a very hard and a very powerful way with me. I left everything to join this team.” A powerful illustration of this can be found in the story of Karen Gripp, who heads up the medical genetics division of Delaware’s A.I. duPont Hospital for Children. Gripp has so far found Face2Gene to be of high value in two difficult diagnostic cases — one (first covered by MIT’s Technology Review) dealing with a child suffering from lateral meningocele syndrome and one with an adolescent suffering from Coffin-Siris Syndrome. Coffin-Siris has an absolutely minuscule level of incidence, with fewer than a hundred cases identified. The Coffin sufferer underwent just the kind of “long journey” Gelbman identified: 16 years, almost the patient’s entire life. FDNA’s tech allowed Gripp to diagnose Coffin-Siris in just minutes.
As revolutionary as Face2Gene has been for the large but clinically fragmented population suffering from rare genetic disorders, its potential stretches far beyond that. In January, an app that uses a related method to diagnose autism — with 86 percent accuracy in early testing — was unveiled at the Consumer Electronics Show. Gelbman sees a huge possibility for expansion along similar lines. “We’ve been working on autism spectrum disorder, so that’s definitely on our road map. We’ve done some research and we’re still doing research on a number of cancer disorders, so oncology is definitely on our road map. We’re basically taking a bet on phenotyping being part of the standard toolkit for evaluating medical conditions alongside with genomic sequencing. I also believe that individuals will be phenotyped at every single point of contact with physicians with our technology. This is going to go back and tie into their medical record or their sequencing file — and at every single point in time, because the phenotype changes over time. There’s a longitudinal aspect of this, and progression. Our technology is going to be used to evaluate different medical conditions.”
So this is clearly just the beginning for FDNA. As deep learning and machine sight improve — and as more medical professionals adopt the technology, thus expanding its database — the sky looks to be the limit. In the age of Theranos, technologies like Face2Gene that legitimately improve quality of and access to diagnostic care are all the more important.