By KEN LEISER
Researchers at Barrow Neurological Institute in Phoenix are using artificial intelligence to better understand the origins and potential treatment of amyotrophic lateral sclerosis, or ALS.
Dr. Robert Bowser is chairman of neurobiology and director of the ALS Research Center at Barrow, which is part of Dignity Health St. Joseph's Hospital and Medical Center. In a conversation with Catholic Health World, Bowser expressed measured optimism about the potential for artificial intelligence to advance ALS research.
![w190615_BarrowResearchers-Bowser_a-1 w190615_BarrowResearchers-Bowser_a-1](/images/default-source/catholic-health-world/w190615_barrowresearchers-bowser_a-1.jpg?sfvrsn=37e0fcf2_2)
Bowser
His research team used artificial intelligence tools to discover five new genes associated with the fatal neurological disorder. The team employed an IBM Watson Health technology platform that applies advanced natural language process, machine learning and predictive analytics to advance research in cell biology. Scientists hope the addition of AI will speed the development of new ALS therapies.
Separately, Barrow researchers are using AI to detect subtle changes in speech patterns of ALS patients (an effort that may offer insights into the biology of disease progression) and they are using AI in an effort to predict treatment outcomes using electronic medical records.
What is the state of medical research into treatments or potential cures for ALS?
It's an exciting time because we've made a lot of progress and inroads in the last five to 10 years, but we still don't have effective therapies to treat these patients with a horrible, hideous disease. But as science is evolving and our ability to ask certain questions has evolved, we've made incredible insights and discoveries into both the basic science and actually into the disease itself.
In 2016, you teamed with IBM Watson Health and used an artificial intelligence algorithm to identify five new genes that are linked to ALS. How did that work?
IBM Watson used its text-based algorithms and a semantic approach to review published literature, including abstracts of previously known RNA-binding proteins. It made interpretations and then tried to almost make guesses as to what's the next leap and what can be predicted based on that.
More than 30 genes have been linked to ALS, and mutations in the 11 genes that encode RNA-binding proteins cause familial forms of ALS. A person's DNA encodes for more than 1,500 RNA-binding proteins. It is unknown if other RNA-binding proteins may contribute to ALS. With so many, the cost and time required to examine them all would have been prohibitive.
The Barrow team provided a list of 11 RNA-binding proteins with known mutations that cause ALS. Watson cross-referenced the list with medical literature from 28 million MEDLINE abstracts to rank order all other 1,500 RNA-binding proteins encoded by our genome to try to identify those linked to ALS. What the Watson platform couldn't do was take experimental results from my scientific lab and upload that information, and then use that information in combination with the text-mining approaches to gain new insight.
How would you improve upon artificial intelligence research tools?
I think the most obvious way we need to go is to be able to incorporate findings from the actual benchwork, from the actual lab itself into the artificial intelligence database query. Uploading that information into an algorithm is a challenge due to the various types of platforms that generate experimental data.
In the Barrow research project ALS AT HOME, patients who are early in the disease provide weekly speech samples by phone or tablet. What do scientists hope to learn from this research?
There are a few other groups around the country using these types of speech or voice analytics with AI approaches to look at speech and speech recognition in ALS patients. The idea is to determine if we can identify changes that either predict the course of the disease, predict the future events that might occur clinically in a disease, or use speech as an outcome measure in clinical trials. So, if you have a drug that would be impacting the neurons that help control speech, if you're protecting those neurons, then the deficits in speech should slow or level off if the drug is effective. Could that be an outcome measure that the FDA would find enticing to actually approve a drug? That's down the road, but that's where some of that research is moving. That is very exciting.
Does the algorithm recognize changes in speech that are so subtle as to be undetectable by the patient or their loved ones?
Yes. The analytic tools could separate (patients with ALS) rather quickly from the control subjects. But it also separated ALS patients that self-reported no abnormalities in speech from the control subjects, indicating that the algorithm could detect speech alterations before the patient or the patient's family could recognize any changes in speech. That again is very exciting. As we are looking for tools or biomarkers for the disease, right now what we use are obviously the hard-core clinical assessments. Those clinical assessments are global assessments of the patient. In order to impact and change those global assessments, a lot of underlying biology has probably changed internally within the patient. Finding biomarkers that are able to quantify those underlying biological changes and therefore distinguish changes in ALS patients prior to our current clinical parameters are greatly needed to help us identify more effective therapies. This AI project is one approach to doing just that.
Do you see artificial intelligence playing a significant role in the future research into ALS?
Yes. I remain hopeful that AI approaches will provide an impact toward the development of new treatments, new ways to monitor patients, new outcomes measures for trials. I am hopeful that AI approaches will lead us in that direction and be a part of that solution. We are still in early stages of incorporating AI into ALS or any other diseases for that matter, but I see a great need and potential to use AI in this disease and I think we need to leave no stone unturned. AI provides a new opportunity to make new discoveries and make new insights into the disease.