Omar Abdul-Rahman, MD, Director of Genetic Medicine and Friedland Professor at MMI, is the senior author on an article recently published online by the American Journal of Medical Genetics – Part A.1 The paper reports his study team’s application of a “deep learning” facial analysis technology to the diagnosis of various forms of the genetic condition Angelman syndrome. Dr. Abdul-Rahman graciously agreed to respond to a short Q&A interview about this work for the CHRI Research Bulletin.
Q: How does Angelman syndrome affect children who inherit it?
A: Angelman syndrome is a neurodevelopmental disorder that causes microcephaly, intellectual disability, little or absent speech development, seizures, and ataxia.
Q: How many different subtypes of Angelman syndrome exist, and what are the usual ways of diagnosing them?
A: There are four primary subtypes based on the molecular cause. Fundamentally, it is due to a reduction in level of the gene UBE3A on chromosome 15 that is expressed in the cells. You can have this from a deletion of chromosome 15 that involves the gene, when both copies of chromosome 15 originate from the father (instead of one chromosome from each parent), when the gene is silenced, or when there is a mutation in the gene. We have historically known that the deletion tends to be more severe because it can impact other genes nearby. But without using genetic testing, it is hard to determine what subtype a patient has. And knowing the subtype that a patient has is important because the prognosis and the chance of this happening again in a future pregnancy are dependent on that information.
Q: What prompted you and your collaborators to explore the use of facial analysis and deep learning to differentiate Angelman syndrome subtypes?
A: There have been studies of some other syndromes using facial analysis and deep learning showing the ability of the system to detect differences in the facial features that are below a clinician’s ability to identify. We also knew that the deletion form of Angelman syndrome tended to have more severe features, so we surmised that there may be differences in the facial phenotype as well that have historically been missed on a clinical exam.
Q: Through this study, what did you learn about the predictive power of this technology for improving diagnosis and care of patients?
A: Although genetic testing has been getting cheaper, there can still be issues related to access to testing. For example, insurance payors are getting tighter about when they will authorize testing, and rural areas especially in the developing world have difficulty in getting samples to a laboratory that can conduct the test. So we were pleasantly surprised when we discovered the technology could recognize the differences between each of the subtypes fairly well. This suggests we may have the ability to bypass genetic testing to predict the subtype using this technology, which is easily accessible on a smartphone and can be deployed virtually anywhere in the world.
Q: Whom did you work with to carry out this work? Are there plans to study it further in the future?
A: Our lead author, Diego Gomez, is a senior undergraduate neuroscience student at Creighton, and we had two other key collaborators. Dr. Lynne Byrd at the University of California San Diego oversees the Angelman syndrome rare disease consortium and we leveraged their extensive database of patients. And Nicole Fleischer was our collaborator at FDNA, the company that developed the facial analysis system. We hope this type of study can be extended to many other syndromes that have more than one underlying etiology to see if there are differences the system can detect. This will help advance our understanding of those conditions, and hopefully further develop this tool that can be used by healthcare providers in the field.
- Gomez DA, Bird LM, Fleischer N, Abdul-Rahman OA. Differentiating molecular etiologies of Angelman syndrome through facial phenotyping using deep learning. American Journal of Medical Genetics - Part A. 2020 Jun 11. doi: 10.1002/ajmg.a.61720. E-published ahead of print. https://onlinelibrary.wiley.com/doi/abs/10.1002/ajmg.a.61720