The Center for Data Innovation spoke with Dekel Gelbman, chief executive officer of FDNA, an artificial intelligence and genomics startup based in Boston. Gelbman discussed how big data can help combat rare diseases and how facial analysis can help doctors better diagnose diseases.
Joshua New: FDNA is both a genomics company and an artificial intelligence company. Can you explain how these overlap?
Dekel Gelbman: Artificial intelligence and machine learning are very powerful tools when it comes to processing huge amounts of data and detecting patterns and correlations that could be scientifically meaningful. Genomics is one of those areas where the sheer amount of data is the biggest hurdle for adoption, specifically when it comes to applications for precision medicine and rare diseases.
Our technologies help reveal these patterns and connect the dots between clinical signs of patients and their underlying genetic causes. This is extremely important for diagnostic tools and for drug developers working on genomic-based medical solutions. With our tools, they can hone in on the exact genetic cause of a medical condition.
New: Why the focus on rare diseases? What makes a company like FDNA better suited to tackle the problem of rare diseases than, say, a pharmaceutical company or research institution?
Gelbman: Rare diseases have a huge unmet need. When you look at the number of patients with rare diseases, as a group, rare diseases are actually quite common. There are 30 million people in the United States alone with rare diseases—that’s about one in ten people, twice the number of patients living with a cancer diagnosis! Also, it’s mostly children that suffer from these devastating diseases. Rare disease patients and their families spend seven years and see seven doctors, on average, before reaching a diagnosis, if they’re lucky enough to even reach one.
One of the most significant challenges in this field, which we are uniquely positioned to overcome, is aggregation of sufficient data from all over the world to understand more than 8,000 diseases. With the largest network of genomic clinicians, labs, and researchers in the world, we are able to crowdsource data otherwise not available and facilitate research.
That’s why drug developers partner with us. We help them discover new biomarkers that help them understand the genomics behind diseases, design end-points, and recruit patients for clinical trials and find relevant patients that can benefit from their treatments.
New: FDNA developed a suite of apps called Face2Gene that use facial analysis to help diagnose diseases. How does this work? How reliable is it compared to traditional diagnostic methods?
Gelbman: Thousands of genetic disorders present distinct facial characteristics. Down syndrome would be a good example of that. Clinicians have been evaluating facial features for decades to help them reach a clinical diagnosis of genetic syndromes and ordering appropriate genetic testing to confirm them. We have built the most comprehensive and largest database of data representing facial images of patients with genetic disorders. We use deep learning to extract unique patterns from those facial images that correlate not only to a disease, but also to the exact genetic cause of that disease. When a doctor uses Face2Gene, we analyze the patient photo and return a list of diseases from our database that can fit those patterns.
Face2Gene itself is not a diagnostic tool. It allows clinicians to reach pertinent information faster and more accurately and assists them when examining a patient. Recently, with genome sequencing becoming the preferred method of genetic testing, we made our technology available to genetic testing labs as well. With Face2Gene used both by clinicians and labs, they are now able to improve the efficacy of diagnostic tests significantly—this means that molecular tests could become even more affordable and accurate and open up the door for other precision medicine applications.
New: FDNA combines genetic, health, and biometric data. Are there any kinds of data that would be valuable in helping study and treat rare diseases that you cannot get access to? What are the obstacles?
Gelbman: Genomic data has two components: genotype information; the actual DNA sequence, coded in A,C,G, and T; and phenotype information, the observable characteristics of an individual resulting from their genetic sequence. The phenotype information, which is our main focus, can be obtained from multiple types of sensors—photos, medical images, voice, vital signs, and so on. There is also a breadth of phenotype information included in doctors’ notes.
Our initial focus was facial photos, since they are extremely informative for rare diseases, but we are now working on collecting and studying additional biometric data to expand our ability to analyze phenotypic data. As with facial photos, obtaining this scarce data for rare disease patients is a challenge, but we already have the largest network of genomics professionals, giving us a huge advantage.
New: Can you explain FDNA’s Centers of Research Excellence program?
Gelbman: Our large network of genetic professionals is extremely helpful for collecting data and understanding rare diseases from all over the world. In addition to this approach, we have decided to take a more proactive research approach and partner with leading genetic institutions around the world to address more specific topics, such as understanding the genetics of certain communities based on their ethnicities or geographic location, or investigating more deeply a specific group of diseases.