Increasing diagnoses will help us cure SLC6A1
SLC6A1 and other genetic brain disorders are underdiagnosed due to barriers to genetic testing. Conditions that are considered ‘rare’ struggle to attract medical research. A great deal of medical research is funded by drug manufacturers seeking a market for their treatment products. Larger market opportunities attract more funding for research than the small markets from rare diseases. The market opportunity for treating genetic brain diseases is larger than we currently know. We can access more funding for research if we can increase the count of diagnoses to represent the actual opportunity for drug and other treatment options.
Most SLC6A1 patients have a long history of EEG testing even while they face barriers to genetic testing. We can add artificial intelligence (AI) researchers to our genetic brain disorder research community by creating an anonymized public dataset of EEG test results labeled with genetic diagnoses. Recent AI research on EEG datasets has found diagnostic breakthroughs for brain disorders such as schizophrenia1. We can access this growing AI research community to discover new diagnostic biomarkers for SLC6A1 and similar genetic disorders. Medical doctors armed with a diagnosis prediction from a readily available EEG will have an easier time accessing genetic testing for their patients.
Barriers to genetic testing
Patients and their families living with debilitating brain disorders are discovering genetic causes of their conditions. The journey to a definitive genetic diagnosis takes years for many families. Many patients never receive an available diagnosis due to hurdles in the availability and costs of genetic testing.
During the journey to my daughter Katrina’s diagnosis at age 13, we watched her fall behind typically developing children of her age. She faced diagnoses of autism, epilepsy, and intellectual disability. Her occasionally extreme behavioral challenges led to injuries and police encounters for our family and her support community of teachers and therapists. After years of working with accomplished pediatric neurologists in the San Francisco Bay Area, we self-referred Katrina to an MD PhD Geneticist at UC San Francisco Medical Center in late 2019. We waited 6 months for her appointment and another 6 months for the genetic test results prescribed by the geneticist. Finally, in the fall of 2020, we received the diagnosis of a defect in one of Katrina’s SLC6A1 gene pair. The SLC6A1 gene was discovered in 2015 and genetic testing panels began diagnosing the disorder in 2017.
The news was hard for our family but it was also our great fortune to find the SLC6A1 Connect community founded by Amber Freed. This amazing community has ended the isolation we felt as we spent years trying to help Katrina on our own. While many of the children in the SLC6A1 community were diagnosed as babies, I have met several other families, like our own, that received their diagnoses when their children were teenagers or young adults.
Many of these families had a diagnostic journey like we experienced with Katrina. They had to self-refer to many different doctors until they found their way to a genetic testing panel that included SLC6A1. Some were prescribed genetic testing only to then face a denial of coverage from their health insurance provider due to the test not being ‘medically necessary’. Families had to pay out of pocket for testing. Free genetic testing from programs such as Invitae’s Behind the Seizure and Probably Genetic were not widely known and had restrictions. Some overwhelmed families give up at this stage and do not receive a diagnosis that is readily available. All of these patients have a history of EEG testing before they reached this point.
Katrina was fortunate. I was self-employed in 2019 and I purchased an Affordable Care Act (ACA) health plan for my family. Our Blue Shield ACA plan covered the costs of Katrina’s testing. Beginning in 2020, I spent over two years as an employee of Google. Google has an internal community of parents of disabled children. This is a wonderful community and I helped advise parents of children with issues similar to Katrina’s to seek genetic testing. To our dismay, many of these families were denied insurance coverage of the testing, despite having coverage from one of the most sought-after employer health plans in the United States. This disparity helps illustrate the variability in access to genetic testing that exists even in wealthy countries like the United States. In the developing world of countries, it is even harder to access genetic testing. While volunteering at the American Epilepsy Society (AES) conference last year, I spoke with physicians from Central and South American countries who do not have access to genetic testing for most of their patients. I’m sharing these stories to illustrate that there are communities of people worldwide with debilitating genetic brain disorders who are going undiagnosed.
It is also clear that we are not reaching adults who have suffered these conditions throughout their lives. The SLC6A1 gene was only discovered in 2015. How many adults living today with epilepsy and intellectual disabilities have undiagnosed SLC6A1-related disorders? I spoke with many physicians at last year’s AES conference. A few of them said they didn’t have SLC6A1 patients because their practice focused on adults. I tried to advocate for genetic testing anyway. The diagnosis is still considered new and, without an ICD-10 diagnostic code for SLC6A1, many doctors are not looking for it.
Accessing AI research
Continuing advances in Artificial Intelligence can be enlisted in this diagnostic effort. Artificial Intelligence algorithms find patterns in data. A great deal of research goes into these algorithms but the research requires large amounts of data to train and test the algorithms. This leaves the AI community hungry for new data for their research. I was in the Product Development organization at Netflix that released the Netflix Prize2 dataset in 2006. The Netflix Prize dataset consisted of 100 million moving ratings from 480,000 anonymized users. Any researcher who could beat the Netflix movie recommendation system by 10% won a cash prize. The fact that more than 20,000 researchers signed up shows the size of the AI research community waiting in the wings to help us. The Netflix prize was won in 2009.
Recent years have seen the public release of EEG (Electroencephalography3) data sets that have also attracted a great deal of AI research. EEG data is attractive to AI researchers because of the large volume of data in even one EEG. Each electrode pair of an EEG is represented as a channel. There are 20 to 25 channels in most EEGs. Each channel records electrical signals at a high sample rate. One 30-minute EEG of 20 channels sampled at 250hz results in 9 million unique data points.
One popular EEG dataset is the Temple University EEG Resource project4, led by Professor Joseph Picone. Over 8,000 researchers have subscribed to the EEG datasets from this project. One dataset from Temple is known as Temple University Abnormal EEG Corpus (TUAB). The TUAB dataset consists of 3000 EEGs. Each of the EEGs has been labeled as either normal or abnormal by a team of neurologists trained to evaluate EEGs. Several research papers have been published in recent years about AI programs trained on TUAB data to achieve highly accurate detection of abnormal EEGs.
There is other recent research that has made diagnostic breakthroughs using much smaller EEG datasets. The Schizonet paper research from 2023 is based on just 193 unique patients of which 85 have a labeled diagnosis of Schizophrenia. The success of Schizonet on small numbers of patients gives me hope that we can assemble a meaningful dataset of abnormal EEGs labeled as Positive or Negative for genetic disorders.
SLC6A1 Connect proposes to develop a new EEG dataset in partnership with academia and other research organizations with access to patient EEG data and the genetic testing results for those same patients. A good first dataset would consist of anonymized abnormal EEGs, formatted in an open standard such as EDF (European Data Format), and labeled with as Positive or Negative for genetic disorders. We know of no such existing public dataset at this time. The goal is to spur AI research into automated systems that would lead doctors to prescribe definitive genetic testing for patients with EEGs containing Positive biomarkers for genetic disorders. The resulting testing would lead to an increase in diagnoses for genetic disorders like SLC6A1.
Kevin McEntee (firstname.lastname@example.org) is available to advise on the project. Kevin is an SLC6A1 parent who wrote this blog post. Kevin was a member of the Netflix engineering team that released the Netflix Prize dataset and continues to engineer AI data systems in the Climate Tech space at Equilibrium Energy.
- SchizoNET: a robust and accurate Margenau-Hill time-frequency distribution-based deep neural network model for schizophrenia detection using EEG signals
- The Netflix Prize (https://en.wikipedia.org/wiki/Netflix_Prize) ↩︎
- Electroencephalography (https://en.wikipedia.org/wiki/Electroencephalography)
- Temple EEG Project (https://isip.piconepress.com/projects/tuh_eeg/index.shtml)