Anastasia Razdaibiedina: Discovering gene-disease relationships with deep learning

The next T-CAIREM Trainee Rounds on August 10  will feature a presentation by Anastasia Razdaibiedina, a PhD student in Computational Biology and Machine Learning at the University of Toronto. She'll discuss her research into using deep learning to discover gene-disease relationships. We caught up with her to learn more,

What inspired you to pursue the field you’re majoring in?
My parents are physicians, so I've always been captivated by biology and medicine. On the other hand, I also loved computer science and math since high school. I majored in applied math and gained research experience in computational biology during my undergraduate years, and have wanted to work in the field ever since. 

How do you explain your research to those of us with Arts degrees? 
Most diseases have a complex genetic basis, i.e. they are caused by a group of genes. How can we discover such groups of disease-associated genes? First, we can find some sort of similarity score between all genes (I will discuss the score in more detail shortly). Based on this similarity score, we can build a network of all gene-to-gene associations, with more "similar" genes being closer in the network. Then, we can annotate known genes with their disease associations. Some genes will form dense neighbourhoods. If, for example, we have a neighbourhood of 30 closely related genes, and 29 of those genes are related to Autism spectrum disorders, we can conclude that the last gene is also related to Autism. This approach allows us to map genes to their associated diseases. Now back to similarity score - it can come from various sources of data (sequence similarity, genetic interactions, protein-protein interactions, etc.) In my case, I am using microscopy data coupled with deep learning to derive the similarity scores. This new approach allows us to investigate the problem from a different angle, and also contains more data than most existing data sources. 

What inspired you to research this topic?
During my undergraduate research, I understood that despite increasing amounts of data being generated, our understanding of gene function and gene-disease relationships remains incomplete. And I was fascinated by how emerging AI methods (trained on millions of data points) are outperforming existing approaches in many areas of biology, or reaching physician-level accuracy in disease identification. These factors inspired me to focus on applying machine learning to solve current problems in biology and medicine.

What outcomes do you hope your research will eventually lead to?
The approach I'm currently developing is capable of characterizing gene function and calling gene-disease hubs. We have extensively validated the method on well-studied yeast data, and are now working with human cell data. Our approach can shed light on unknown gene function across different organisms and improve understanding of disease mechanisms. 

What are your professional goals after you complete your studies at the U of T?
I'd like to continue the research route. I would be thrilled to develop my current research ideas to expand our understanding of gene function and gene-disease relationships on many more types of biological data. I also want to explore other areas of machine learning and their applications to medicine and biology.

What do you like to do when you aren’t working?
I like doing all kinds of sports - running, hiking, skiing, playing tennis and yoga. 

What advice would you give to high school students who are interested in following in your footsteps?
Let your interests drive your studies/career and keep exploring! A lot of exciting research is happening at the intersection of different disciplines, so you don't have to specialize in one field. Keep the doors open and don't be afraid of changes and experiments.

What’s the best part of doing the type of research that you do?
The best part is being able to develop and apply the latest machine learning algorithms to solve real-world problems. Knowing that methods that I am developing today could be used in a hospital or research lab tomorrow is very motivating. I am excited to bring cutting-edge AI research into clinical settings.