NEW YORK – A new partnership involving researchers at Stanford University, Kailos Genetics, and digital health firm Doc.ai could result in new models for using pharmacogenomics data and artificial intelligence to predict the best treatments for patients with epilepsy.
Led by Robert Fisher, director of the Stanford Epilepsy Center, the Epilepsy Digital Health Trial commenced last month with the aim of combining personal information, pharmacogenomics data, and machine-learning tools to select the best anti-seizure medication for patients.
While this trial is focused on epilepsy, the commercial partners of the collaboration believe they could apply their methods in future studies focused on predicting the best therapies for patients with other diseases.
"Trial and error with medication is an issue for many diseases," said Samantha De Brouwer, COO and cofounder at Palo Alto-based Doc.ai. "There is this idea of accelerating what we can understand by combining omics with people's participation," she said. "If we can build models for epilepsy, that would give us hope for other types of diseases."
According to Fisher, the Epilepsy Digital Health Trial has its roots in a desire to use genomics to help identify the optimal treatment for different patients. Currently, there are around 25 therapies for epilepsy, but some patients react poorly to some medications and better to others.
"Every epilepsy drug is processed by enzymes, which are coded by genes," said Fisher. "We know of several examples where genomics affects efficacy or predisposition to side effects, but there must be many more," he said. "Our hypothesis is that machine learning will be able to identify predictive patterns for individual reactions, both beneficial and harmful, to antiepileptic drugs."
To achieve that, Fisher's team turned to Doc.ai, a three-year-old company that relies on blockchain-based AI to produce predictive analytics and personal health information. According to De Brouwer, doc.ai does not use a single underlying AI technology, but employs a suite of technologies based on what it aims to accomplish for the specific healthcare challenge.
The company, which employs 50 people, has embarked on two similar trials so far — one focused on food allergies, with partners at Harvard University, and the other on irritable bowel syndrome and Crohn's disease, led by investigators at Kansas Medical Center. The epilepsy trial seeks to enroll up to 1,000 participants between September 2019 and September 2020. Participants have to be between 18 and 100 years old, have an epilepsy diagnosis, have a smartphone using the iOS operating system, and live in the US, according to the partners.
Patients will use a Doc.ai app to record seizure episodes and side effects related to medication for three months. They will also collect information about their physical traits, exercise and activities, other medications they are on, and their environment. Each patient will be tracked for a three-month period, at the end of which they will receive back an individualized report about data collected via the app that they can share with their clinicians.
A core component of the trial will be the genetics data obtained via Huntsville, Alabama-based Kailos Genetics. CSO Troy Moore said that Kailos modified its next-generation sequencing assay to cover targets of interest to Fisher's team at Stanford. Kailos' approach relies on its non-PCR-based enrichment method, called TargetRich, combined with sequencing. The company has been deploying its approach in pharmacogenomics research for years.
"We went back and forth a lot ... about what the appropriate markers are for this study," Moore said. "We used our current assay as a basis, made sure that certain markers were included, and made sure they aligned with medications that are reasonable and expected to be utilized for treatment for epilepsy patients," he said.
Moore noted that Kailos has the ability to refine and fine tune the assay as time goes on, because it is developed in house. "We have ultimate control over it and the flexibility to change it if appropriate," he said.
According to De Brouwer, it was Doc.ai that reached out to Kailos to contribute to the study. "When they started to tell us what data they would like to collect, genetics was one of them, so we were looking for a strong partner, a very qualitative test, to be able to fulfill the research," she said. "Kailos was interested in our approach and we were interested in collaborating."
De Brouwer noted that the work with Stanford and Kailos is very much in line with what the company had been seeking to achieve since its founding three years ago.
"From the high level, what we have been trying to do at Doc.ai is to accelerate research, to develop tools and modules that facilitate the organization of trials," she said. "But our company also wants to broaden the scope of what is possible today in clinical trials, and enable the collection of different types of data."
"There is a lot of trial and error when it comes to cocktails being given to patients with epilepsy and they don't understand what is working, what is not working, on top of the side effects," De Brouwer went on. "The point is to use machine learning to enable the permutation that humans cannot do, and to support different kinds of data collection on top of other surveys for patients."
According to Chethan Sarabu, director of clinical informatics at Doc.ai, his company will work with the vast amount of data collected via the trial to see if its machine-learning tools can determine the best ways to treat patients.
"The ultimate question we are trying to figure out from the study is, 'Can AI predict the optimal epileptic drug?'" Sarabu said. "That is very much a permutation problem that is well defined because there are 25 medications right now to treat epilepsy and we know that there are certain factors that go into which medications may work for certain people."
The investigative team believes that some combination of the information will point the right way, he noted. The goal is "to see if AI can use all this data to find permutations that provide similar answers and clarity around which medications work best for which people," he said.
It's the first time Doc.ai has included genetic data in a trial, which Sarabu said is an "exciting evolution" of the firm's platform and showcases that the firm can bring together omics data and real-world data, much of which can be collected via mobile phone in a person's home. "I think it's an exciting culmination of the work that has been going on for some time and an expansion of our trials platform," he said.
Sarabu noted that the first results from the trial should be expected sometime next year.
Kailos' Moore said the company was interested in taking part in the new trial because the company seeks to bring genetics to patients. "That was one thing that attracted us about working with Doc.ai, the direct patient involvement in the study," Moore said. He also for future collaboration.
"We are always evaluating different approaches, and certainly the way Doc.ai was structuring this study was very interesting and groundbreaking and something that could be a model, not just for epilepsy but for other studies going forward," said Moore. "That's why we were intrigued by it."
He noted the firm continues to be interested in leveraging its knowhow and platform to "enable different groups that are starting to push different ideas for performing clinical studies."