Email
Password
Remember meForgot password?
    Log in with Twitter

article imageQ&A:Transforming genetic medicine as the medical standard of care Special

By Tim Sandle     Mar 19, 2020 in Science
Can genetic medicine become the medical standard of care? Scientific developments are heading this way. One example is a new machine learning platform from the biotechnology firm Emedgene. The company's founder explains the technology.
With rare diseases, 72 percent out of the 7000 known are genetic, and 70 percent of those start in childhood. The lack of scientific knowledge and the quality of information often delay diagnosis or lead to misdiagnosed cases, losing precious time that can be vital to find treatment before it's too late.
This situation is changing with the advent of genetic medicine. an example is Emedgene's artificial intelligence software, which is the world’s first completely automated genetic interpretation platform using machine learning algorithms.
Digital Journal spoke with Einat Metzer, CEO and Co-Founder of Emedgene to talk about the new genetic interpretation software.
Digital Journal: How are rare diseases classified?
Einat Metzer: Rare diseases defined by the number of people affected. In the U.S., any disease that affects fewer than 200,000 people is defined as rare, in Europe, it’s any disease affecting fewer than 1 in 2000 people.
There are around 6000 known rare diseases, and that number is growing. What’s interesting to know, is that although they are each individually rare, collectively they impact over 300 million people. Those patients have a very difficult time receiving a diagnosis for their disease, and typically go through a diagnostic odyssey lasting on average 5-7 years. It’s also worth noting that most rare diseases have a genetic basis, and appear in early childhood.
DJ: Is sufficient funding and research invested into rare diseases? What are the factors that influence this?
Metzer: There are two challenging aspects to rare diseases, the first is the identification of a rare disease, because obviously, physicians aren’t familiar with every disease affecting only tens or hundreds of patients worldwide. The second difficult aspect is developing treatments for diseases impacting small numbers of patients.
The good news is countries and healthcare systems are increasingly recognizing the need to cover genetic testing for the identification of rare diseases. As of today, over 50% of the US population has insurance coverage for next generation sequencing.
However, even insurance coverage for the tests does not entirely solve the problem. Sequencing a patient’s DNA is easily done, but understanding what variants in a patient’s genome mean is still quite challenging. Every patient has millions of harmless genetic variants, and only one disease-causing mutation. As a result, geneticists can spend hours manually reviewing hundreds of variants and looking for evidence for the disease in databases and the literature.
There are fewer than 5,000 geneticists worldwide available to interpret patients’ genetic data, resulting in an interpretation bottleneck. Even as more patients become eligible for genetic testing, the workforce capable of diagnosing them is not growing fast enough. We estimate the worldwide capacity of interpretation is capped at roughly 2.4 million tests, less than the predicted rare disease testing volume for 2020.
DJ: How can machine learning help?
Metzer:Machine learning technologies can reduce the manual labor of interpretation, by offloading both the research and deep analysis tasks from geneticists. Machine learning is a buzzword, widely used, and applied to many types of solutions. We’re talking about a unique application of the technology here, where we won’t use a single algorithm to solve a single problem. Instead, we need to apply a set of algorithms designed to automate different aspects of the geneticist’s workflow.
On the one hand, the geneticist’s work is to review thousands of data points for every patient’s test, and use them to come to a conclusion on the single genetic variant that’s causing the disease. We can certainly apply machine learning algorithms to review those data points. But we can go a step further, and collect the data points most likely to impact their decision, and include those in our recommendations.
The second labor-intensive task geneticists perform, is looking for the most up-to-date information in databases and the published literature. That’s a task well suited for Natural Language Processing, which can be used to augment existing databases with information curated from the literature.
DJ: How does Emedgene’s AI software work?
Metzer:Emedgene’s AI-powered genomic analysis platform tries to do just that, automate the labor-intensive parts of the geneticist’s workflow, so interpreting a patient’s genetic test takes less time and effort, and accuracy is not compromised. The goal is to scale the genetic testing interpretation in healthcare systems, so they can offer personalized care to a broader population.
Our AI consists of dozens of different algorithms, each solving a different problem, all coming together to automate the genetic testing interpretation workflow.
The platform is able to automatically identify the disease-causing variant, compile the evidence, and present it to the geneticist on the case for review. The machine learning algorithms utilize a proprietary knowledge graph that continuously incorporates new knowledge. The knowledge graph contains over 85,000 entities and 340,000 connections today, including unique information curated from the literature that has not yet made its way into public databases.
DJ: What were the main challenges when developing the software?
Access to large high-quality data sets is a major challenge in developing AI solutions in healthcare in general.
For our supervised learning algorithms - those that require labeled data for training the algorithm - once we obtained the data, labeling was a challenge as well. The level of education required to annotate healthcare datasets is quite high.
Fortunately, there are good solutions to both problems, both from the scientific and AI perspective.
DJ: Are there any case studies you can share, to show the benefits of the approach?
Metzer:We’ve studied the accuracy of our interpretation algorithms with Baylor Genetics. In the 180-case study, our AI successfully identified the disease-causing mutation in 96% of the cases.
Another of our customers, Greenwood Genetic Center, was able to reduce time spent per case by 75%, which was translated directly into shorter turn around times for patients waiting for a genetic diagnosis.
More about genetic medicine, Medicine, Medical, Genes
More news from
Latest News
Top News