Mar 26, 2021

How experimenting with machine learning changed the future for a PhD candidate

Programs: Graduate, Research: Artificial Intelligence in healthcare, Research: Brain & Neuroscience
Joey Silburt

Joey Silburt is in the final stages of completing his PhD in the Department of Laboratory Medicine & Pathobiology in the Temerty Faculty of Medicine. Having a fascination for gene therapy, he joined Dr. Isabelle Aubert’s lab. The Aubert lab specializes in using therapeutics that can be delivered to the brain with focused ultrasound and eventually help people with Alzheimer's disease.

“One of our biggest problems is actually getting therapeutics into the brain,” explains Joey. “It’s a very delicate, isolated organ that protects itself very well via the blood-brain barrier.”

Dr. Aubert’s lab has been using a transcranial focused ultrasound to localize acoustic energy and cause a brief increase in the permeability of the blood-brain barrier, allowing substances from the blood to leak into the targeted brain regions. This can then be used to deliver therapies from the blood to the brain in a non-invasive way, as opposed to inserting needles containing the drugs in the brain tissue, which can cause a lot of damage.

Working collaboratively with his classmates, Joey started his PhD on the gene therapies that could be delivered using focused ultrasound. While learning the field, he rapidly gained interest on what else could focused ultrasound be impacting in the brain.

Understanding what is (or isn’t) trauma in the brain

Microglial cells are part of the brain’s immune response and are the first to respond when something goes wrong in the brain. They are activated during trauma, for example, a stroke or concussion, and they repair the blood-brain barrier, calm the environment down, and essentially ‘clean up’ anything that shouldn’t be in the brain.

When using focused ultrasound, the blood-brain barrier is temporarily compromised so microglia are activated and present.

“Some would observe this microglial activation and see it as indicative of trauma,” explains Joey, “This was putting off many from adopting the technique, but I wasn’t convinced we were looking at it in a detailed enough way.”

Joey wanted to study the activity of microglial cells and thought machine learning could be a solution. Although he had some programming experience, he had no machine learning knowledge so had to learn quickly.

A fast introduction to machine learning

His brother was completing a PhD in Astrophysics at the University of Toronto and part of a machine learning study group that Joey joined to grasp the basics. He then took some online courses and taught himself machine learning.

“There are many different types of machine learning”, explains Joey, “For example supervised learning is when you effectively tell the program the answers so it can recognize patterns, such as what a benign or cancerous tumor looks like, and then it classifies new samples into one of those two groups. I’m not an expert at classifying these cells so can’t make those kinds of determinations.”

To solve this problem, Joey turned to an unsupervised model called the ‘one class support vector machine’. He fed into the programming what healthy non-activated microglia look like and the program flags anything that doesn’t look like the control sample. Using this technique, he developed a software, called Morphious, which allows him to track microglia activation over time.

What he discovered was that in severe trauma, such as a stroke, microglia are activated for a prolonged period of time - at least a month. With focused ultrasound, the activation was very brief, significantly reducing after 4 days.

“What we realized was that this was not trauma,” explains Joey, “The microglia respond to the fact that some proteins are being released into the blood-brain barrier, but they calm down and reset very quickly. They do not act in the same way to any other injury or trauma in the brain – it’s a more proactive, regenerative response rather than a pro-inflammatory one so it could actually be of benefit.”

His research has enabled the lab to look at cell activation in a more detailed way and will be used in further studies, such as how certain therapeutics could help reduce inflammation following strokes.

A new focus for a PhD

Studying the activation of microglia was originally planned to be a very small part of Joey’s thesis but he had to use machine learning in order to study it in a rigorous way, which took him down a different path and took four years to build the algorithm.

It has also awakened a deep interest in machine learning for Joey. “When I entered my PhD, I had a clear idea of where I was going. Now the world of machine learning has opened up to me and I’ve realized that is where I want to focus my career now. Whether that’s moving into pure data science or applying more machine learning in a biological context, I’m not sure yet, but machine learning will definitely be part of my future and career.”

Interested in Artificial Intelligence and machine learning?

The Temerty Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM) 

Q&A with Bo Wang: why collaboration is essential for the future of AI in healthcare

Research focus: Bo Wang and machine learning in biomedicine

Two LMP students selected for T-CAIREM Trainee Rounds to present their research in AI

Papers published as part of this research

Weber-Adrian D, Heinen S, Silburt J, Noroozian M, Aubert I. The human brain endothelial barrier: transcytosis of AAV9, transduction by AAV2. An Editorial Highlight for "Trafficking of AAV vectors across a model of the blood-brain barrier; a comparative study of transcytosis and transduction using primary human brain endothelial cells. J Neurochem 140 (2): 192-194, 2017. doi: 10.1111/jnc.13898.

Weber-Adrian D, Kofoed RH, Chan JWY, Silburt J, Noroozian Z, Kügler S, Hynynen K, Aubert I. Strategy to enhance transgene expression in proximity of amyloid plaques in a mouse model of Alzheimer’s disease. Theranostics 9(26):8127-8137, 2019.

Weber-Adrian D, Kofoed RH, Silburt J, Noroozian M, Shah K, Burgess A, Rideout S, Kügler S, Hynynen K, Aubert I. Systemic AAV6-synapsin-GFP administration results in lower liver biodistribution, compared to AAV1&2 and AAV9, with neuronal expression following ultrasound-mediated brain delivery. Scientific Reports 11(1), 1934, 2021.

Silburt J, Lipsman N, Aubert I. Disrupting the blood–brain barrier with focused ultrasound: Perspectives on inflammation and regeneration. A Letter to the Editor in response to “Disrupting the blood–brain barrier by focused ultrasound induces sterile inflammation”. Proc Natl Acad Sci USA. 114 (33), E6735-E6736, 2017. Impact Factor 9.6 (Editorial)

Silburt J and Aubert I. MORPHIOUS: machine learning workflow to naively detect the activation of microglia and astrocytes. bioRxiv, 2020.