How can we use machine learning to improve drug discovery, patient treatment, and outcomes? Dr. Simon Eng is putting his PhD to work to answer questions like this and more. We caught up with Simon recently to ask what he’s been up to since finishing his PhD in the Department of Immunology (DoI). Before grad school, Simon completed his BSc at the University of British Columbia studying computer science, microbiology, and immunology. It was during this time that his interest in using machine learning to make sense of biological data first peaked.
I grew E.coli under different colours of light and realized that I could quantify growth by approximating how much of each Petri dish was covered using a mosaic filter in Photoshop. At that point, I realized that the idea of doing both biology and computer science wasn’t farfetched.
While looking into options for his graduate studies, he found out about specific labs and researchers whose work combined computer science and immunology.
I was specifically interested in networks and was fascinated by the parallels between those and the signalling pathways we often mention in immunology.
After finding out that previous DoI graduates came from mathematics backgrounds, he felt his fate was sealed.
I thought, if a mathematician could make a meaningful contribution to the department, then someone with a hybrid computer science/immunology background could too!
Simon’s interests and skills served him well when he joined the labs of Dr. Rae Yeung and Dr. Quaid Morris to work on using machine learning to reclassify childhood arthritis. Suddenly machine learning was providing the main framework for his graduate studies, with immunology providing the foundation for contextualizing the machine learning methods he used. According to Simon, his passions had collided serendipitously, fueled further by the fact that Toronto was quickly becoming the forefront of machine learning research.
Studying two fields as different as computer science and immunology comes with its advantages and disadvantages. The importance of proper science communication was not lost on Simon. While most of us are focused on distilling our projects to an “elevator pitch”, a style of communication that summarizes your work to its most important points, Simon learned to summarize his research while also having to refine topics in a way that would appeal to both audiences.
I had to become good at distilling concepts in each field down to terms that the other side would understand. This partly required some experimentation: how far you could [sic] go in distilling concepts down without losing the core substance of what you are trying to say?
On the other hand, when focusing on multiple angles of research, it’s natural to lose some depth.
No matter how hard you try, your research is either like a black box or its motivations are mind-boggling to some people in either field. It’s hard not to be disappointed, but you learn to move on.
We asked Simon to reflect on his time as part of the UofT immunology department, and how it pre-pared him for the work he does today.
I remain thankful for the colleagues I went through grad school with – talking with them truly forced me to find ways to explain my research in ways that they could understand, which helped me do likewise for non-immunologists.
The DoI opened doors to collaborative efforts with people from a multitude of backgrounds. Currently, he works at BioSymetrics, a company that uses machine learning to integrate clinical and experimental data to better understand human disease pathology. Simon’s work involves collaborating with individuals from multiple disciplines to contextualize human data by using code. At the end of the day, machine learning is a means to an end in the context of research. You have to start with the question you want to ask and use machine learning to build a framework that answers said question.
This happens to be no different from running statistics or generating figures – you would not blindly generate a bar plot or run an ANOVA without knowing what question you want to ask. Models have incredible predictive value, but at the end of the day biology – and by extension, immunology – is far more than just a bunch of probabilities that a model spews out.
If someone was interested in taking a path similar to Simon’s, how would they do it? Well according to him, the most important part is how you start, and this means beginning with a good coding foundation.
R is fine, but Python is great for not having to worry about the intricacies or quirks that others bring. Start small, maybe writing code to help plot simple data like ELISA results. As you start laying the foundations, you can then start to think of projects where you might need machine learning to make sense of the data you’re generating. Great examples include predicting labels or out-comes from data sets too large to interpret by hand, such as RNA-seq data.
Simon is a wonderful example of what can hap-pen when you allow your interests and skills to inform your decisions. And most importantly, that serendipity combined with passion can take you to places you never expected. In Simon’s words, put in the work and “The underlying theory you’ll need will come in due time.” Wise words to live by!
The underlying theory you’ll need will come in due time.
Salma Sheikh-Mohamed
Latest posts by Salma Sheikh-Mohamed (see all)
- Popular Science: Who gets to have an opinion? - June 5, 2023
- The Era of Sustainable Investing: ESG and Corporate Responsibility - January 30, 2023
- Mental Health in the Workplace - April 17, 2022
