The Pheno-Type: Victoria Catterson
Victoria Catterson has been a crucial part of BioSymetrics since 2017. Her journey has taken her through academia at the University of Strathclyde, moving from diagnostics in power equipment into machine learning for healthcare. Read more about her insights below!
“I think that by combining pieces of knowledge from across different domains, we can build systems which are stronger and more innovative overall!”
Tell us more about your journey leading up to BioSymetrics.
In 2017, I was at a crossroads in my career, having just moved to Canada and transitioned out of academia. I was still really excited by the possibilities offered by machine learning and the growing abundance of data, but unsure if power engineering was the place for me. One of my last projects at the University was a collaboration with Biomedical Engineering on patients with Spinal Cord Injury, and I was really inspired by the potential to more directly impact people’s lives. I realized that the same methods could be applied to analyze human disease as machinery failures, but the humans can tell you when they are feeling better! BioSymetrics was the perfect opportunity to apply my existing expertise, while expanding my knowledge of biology, chemistry, and the healthcare industry.
It’s amazing that you’ve been able to make the leap from engineering into biotechnology. What were some hurdles you encountered in making that leap?
One of the biggest challenges was the difference in terminology between the two domains. For example, in engineering, a diagnostic model is any supervised learning technique that predicts a label, and there’s a distinction made between a diagnostic model which predicts the current state of a system, and a prognostic model which predicts the future state of a system. But in biotechnology, you would only call something a diagnostic model if it makes a clinical diagnosis. This is a really tiny class of models which are actually used by medical practitioners in the clinic on the data from a single patient. An engineering diagnostic model is more likely to be called a predictive model in biotechnology, which sounds a bit more like an engineer’s prognostic model.
However, I was surprisingly prepared for this by something else that I worked on before. I have some experience with knowledge elicitation, which is a method of structured interviewing to capture rules and processes around a specific topic. It’s often used to identify business rules or operational protocols before someone retires from a company and the detailed knowledge of how to do their job is lost. Often the steps the person takes are so ingrained that they no longer consciously think about them, and they would struggle to write the process in enough detail that someone else could follow it. The interviewer starts off assuming zero knowledge about the topic, and is able to ask questions in such a way as to tease out this “tacit knowledge”. In the 90s and early 2000s, this was explored as a way of building intelligence into rule-based expert systems.
I approach all my projects from the perspective that others on the team have lots of tacit knowledge that I’m missing, whether that is clinical experience, genomics expertise, or even just information about how a particular dataset was collected. By asking the right questions, I try to sidestep my own terminology biases, and make sure we’re all talking about the same thing, regardless of the precise meaning in other contexts. I think that by combining pieces of knowledge from across different domains, we can build systems which are stronger and more innovative overall!
“I’m excited to see the scientific world continue to gather and analyze these huge datasets, which will eventually explain what’s going on [in health and disease].”
Quite the insight into how we can approach moving between different fields! In terms of the future of biomedicine, what excites you the most about the role of artificial intelligence there?
The biggest thing for me is the scale of the data we have access to these days. We have curated global datasets such as Open Targets and population-scale resources such as the UK BioBank, which can be used either to mine for hypotheses to test, or to supplement data collected for a specific project, and these datasets are so huge that we need statistical techniques and machine learning to sift through them.
One of the most exciting aspects is that these large datasets illuminate how much we still don’t know about health and disease. It used to be thought that mapping the humane genome would explain the majority of disease, but the relative dearth of monogenetic disorders shows that we’re still missing something crucial. Is it epigenetics? MicroRNAs? Something else? I’m excited to see the scientific world continue to gather and analyze these huge datasets, which will eventually explain what’s going on.
Now for something light-hearted! What's something interesting or unique about you that not many people might know about?
One of my hobbies is dressmaking, which I’ve been doing on and off since I was a teenager. I think most people struggle to find off-the-rack clothing that fits well, and I’m no exception. I started making skirts, dresses, pants, and tops, as well as tailoring shop-bought clothes, in order to get the fit I wanted. I used to follow a pattern then tailor the resulting garment, but more recently I’ve started modifying the pattern to fit my proportions before cutting the material. I loved technical drawing in school, and pattern-making is really just technical drawing of the 2D net for a 3D structure!