The face—its angularities hauntingly familiar, as they sit sharply juxtaposed to the supple flexures of skin sprawled across the boney canvas that lies beneath. These features are instantly recognizable as the mask of individuality, while simultaneously presenting as a portal to shared emotional responses. So accustomed have we become to these visages that we search for their reassuring pattern among the mundane backdrops of our daily existence. Faces we contrive within the inanimate makes the cold and industrial seem quirky and approachable, whereas faces among animals or others of our species allow us to be empathetic and bask in their emotions.
As a species, we humans differ genetically by only 0.1% on average. Yet, the outward display of those genes on a grand, 3D scale (what geneticists refer to as phenotype) produces the incredible variety that we visualize every day as we look out from our own unique façade. A face tells a story. The weathered, sun-dried lines of cheeks that have seen decades worth of sunsets after long days spent working outdoors tells one tale. While the rosy, smooth skin that universally makes grandma’s squeal with delight and shout “lemme squeeze that lil’ puddum!” can only be attached to the cherub-like creatures that have just begin to write their story.
But what if a face could tell another story? What if it could accurately, and more importantly, nonverbally, inform all who gaze upon it that advanced medical assistance was required. Now, the vast majority of individuals inherently have the simple investigative powers necessary to discern if something is awry—a sudden change in facial color, dilated pupils, or even the loss of tonality for the muscles of the face. But what about symptoms so obscure that it would be essential to compare the suspected mug to thousands of seemingly healthy faces simultaneously. Regardless of whether any individual beyond Sherlock Homes possesses such extreme powers of observation, the logistics alone would preclude such an experiment from even being possible . . . or at least that used to be the case.
In this digital age, when cell phones and selfies have become a standard term in the common lexicon (perhaps begrudgingly for the later), empathetic entrepreneurs have begun to leverage their technical expertise to create products that are of immense value to scientists and clinicians. Take for example Boston-based FDNA, a company providing technology solutions that are solving problems for rare disease patients in a most unique fashion.
A Rare Find
“As a group, we didn’t come with a predisposition of any genetics background or medical background—we’re technology people,” explained Dekel Gelbman, CEO of FDNA. “In fact, the co-founder and chairman of this company (Moti Shniberg), his previous company was a facial analysis technology company that was sold to Facebook, and they’re now using that technology to tag images. We came with a set of know-how and technology skills, and the mission statement that we started with was how can we apply this technology to the healthcare field and solve a real problem. We did a lot of research and talked to a lot of specialists in all different areas—all the fingers ended up pointing to genetics because that’s how physicians practice dysmorphology.”
Dysmorphology is the study of human congenital malformations—birth defects—in particular, those affecting the anatomy or form (morphology) of the individual. For decades, physicians have tried to master this diagnostic technique, yet until recently they have most often done so without the use of advanced technology. The FDNA team quickly realized they had found a niche where they could leverage their extensive knowledge and maximize their impact—rare diseases.
“Rare diseases as a group are not rare at all,” Gelbman noted. “They affect about one in every ten people on the planet. When you look at one specific disease [in this category], yes, it’s rare, but when you add them, as a group they have a very significant effect, and they affect hundreds of millions of people around the world.”
Gelbman added that “not a lot of people know that this is such a huge problem. There are about seven to eight thousand different rare diseases that have this large effect on the population. This lack of awareness contributes to what we call the diagnostic odyssey, as it takes on average about seven years for a rare disease patient to get a correct diagnosis, during which time they see about seven different specialists.”
To their credit, physicians do their best to assist patients and their families in finding the proper name for their disorder. Yet some of the diseases are so rare that doctors could practice an entire career in a large city hospital and never come across a single case of what some of these patients suffer from.
Fortunately, over the past several years, with rapid declines in price, next-generation sequencing (NGS) has not only allowed many physicians and researchers to be able to identify various rare disorders but has empowered them to determine the specific genetic background (genotype) for an array of rare diseases. Because of this incredible genomic power, dozens of initiatives have begun to enlist thousands of patients with the intent of developing comprehensive genetic maps. However, genotyping is only one-half of the biological characterization process. Harnessing and interpreting phenotypic traits for rare diseases is where FDNA decide to leave its mark.
Putting a Face to the Name
A person’s phenotype is typically the outward result of their genetic code, and information gleaned from observations that utilize phenotypic characterizations help to define patients from a clinical perspective. So, the FDNA team began developing a product that would help clinical researchers and healthcare professionals match up what they were learning from NGS analysis to what they were visualizing from patients features and expressions.
“We wanted to see how we could train a deep learning system to augment the analysis and the evaluation that a clinician does and provide them with real-time access to a library of images to determine similarity in a matter of seconds,” Gelbman remarked. “We got to a point where a physician could upload a regular facial photo, taken with a standard mobile device, and immediately get a list of potential syndromes where the system found similarities in the face.”
After creating a small database of images and training the system to recognize some rare disease variants, FDNA took the next step and created an app called Face2Gene. The app takes a patient’s photo and creates a facial mesh that is compared to syndrome-specific classifiers, or gestalts, and then is quantified to generate a ranked list of possible syndromes based similar morphological data. Since Face2Gene is a learning system, the more data it collects, the better at predicting it gets. Gelbman noted that after they gave the app to a bunch of geneticists to play with, “it was an overnight hit.”
The excitement around the Face2Gene app led the FDNA team to start investigating how they could add even more value to the system. Quickly realizing that the application went far beyond the basics of pattern recognition or facial analysis, these tech experts were following the rabbit hole, which led them to that path of deep phenotyping—often described as the comprehensive analysis of phenotypic abnormalities in which individual components of the phenotype are observed and defined. Since the hope of precision medicine is to provide patients with the best possible care based on disease stratification into subclasses with a common biological basis, the use of deep phenotyping technology such as Face2Gene seems like a big step in the right direction.
“As we connected with more clinicians, researchers, and labs and we began to understand that structured phenotyping is a real challenge for genomics,” stated Gelbman. “Especially as the high throughput technologies like exome sequencing started to develop and transition from research to the clinic, we realized that without accurate, structured, and scalable phenotyping it’s difficult to filter, interpret, and prioritize the variants that are produced from exome sequencing or whole genome sequencing. We were getting approached from the labs and being asked if we could share that information.”
Face2Gene doesn’t just end at analyzing patients faces, however. The app has expanded into a clinical warehouse platform that hosts a suite of five applications designed to aid clinicians, labs, researchers, and educators through approaches such as better variant analysis, enhanced patient evaluation, and even interactive training modules. Gelbman stated that FDNA and Face2Gene are intently focused on making significant contributions to the rare disease field and believes that their technology will play an important part in realizing the full potential of precision medicine initiatives.
