Manoj M Wagle
Manoj is a current PhD student whose research focuses on utilising artificial intelligence and deep learning in biology to advance precision medicine, with a particular interest in neuroscience. His work is based between the School of Mathematics and Statistics on the Camperdown Campus and the Children's Medical Research Institute.
What is the aim of your research project?
My research is about applying transparent AI-based tools in the field of biology – developing interpretable deep learning models for analysing high-throughput biological data, such as single-cell and spatial omics data, for precision medicine.
Further, I am interested in exploring these approaches in the context of neuroscience. Precision medicine is understanding how an individual's unique genetic makeup contributes to their disease and how we can use this valuable information to tailor treatment specifically to each patient with minimal side effects.
Deep learning tools are increasingly being adopted to uncover underlying molecular mechanisms in both healthy and disease conditions. However, a significant problem is that we often don't know how and why these models make certain predictions, which makes them 'black boxes'.
This is especially important in the field of biology and medicine where the reasoning behind predictions becomes crucial. My research aims to address this issue by making deep learning models more explainable and demonstrating how such tools can be utilised to pinpoint complex changes occurring in various diseases.
How and where do you conduct your research?
You will usually find me in the School of Mathematics and Statistics, at the Camperdown Campus or the Children's Medical Research Institute in Westmead, glued to my computer with a cup of tea by my side. My research is purely computational, so yes, no lab coats or test tubes!
I would say it's a blend of biology, coding and cool visualisations. I mainly use the high-performance computing servers (think of these as fancy computers that are hundreds to thousands of times more powerful than your regular one) available in the School of Mathematics and Statistics, as well as resources from the Computational Biology Group at the MIT Computer Science and Artificial Intelligence Laboratory.
What are your findings so far?
Since starting my PhD last year, I have been busy laying the groundwork for my research, during which I have conducted a comprehensive review of various interpretable deep learning-based approaches developed in the field of single-cell omics. Building on that, I have developed a deep generative tool that jointly learns from diverse single-cell molecular data to capture cell type marker genes, that is, what makes each cell unique.
I have applied this model to predict cell types across a wide range of both unimodal and multimodal single-cell datasets and have found it to outperform several state-of-the-art tools. I have also used the markers established by my method from healthy samples to robustly map cell types in our in-house data of early and late stages of Alzheimer’s disease.
What is the most challenging thing about your work?
One of the biggest challenges in my work is making sense of the vast amount of biological data and applying deep learning tools that we can actually understand. I think of deep learning like teaching a computer to learn from your data, and then trying to figure out what is going on under the hood. Making these models interpretable means diving deep into both computer science and biology. So most of the time, I'm working on how I can make my tool work for the data at hand and ensure that it is biologically sensible. While these challenges can be tough, they are also what makes my work exciting!