Grant: Discovery Launch Supplement

Thanks to advancements in data collection techniques, researchers can now gather a lot of measurements about the same subjects from various sources. For instance, in a molecular biology experiment, researchers may obtain data on a specific gene, including DNA sequencing information, gene expression levels, protein interactions, and epigenetic modifications, all from the same biological sample. This kind of measurements, called high-dimensional multi-modal data, need advanced data science methods for processing and analysis. The main aim of my program is to develop easy-to-understand deep learning models to deal with challenges in analyzing these complex data. The challenges include integrating different types of data, exploring differences within the same type of data, selecting important information from the large data to make the data relatively small, examining how the data changes over time, and making the results and analysis more human understandable. To handle the integration of different data types, my team will develop a new technique called generative deep learning model. It can learn the complex relations between different data types and generate one data type from the other data type. In this way, even when some data is missing, we can still make the analysis possible. In addition, the generative deep learning model can generate data without privacy concerns, making it shareable with other researchers. To explore differences within the same type of data, we will develop another advanced deep learning model called graph neural network. We will combine this graph neural network with classical statistical/mathematical models to examine how the data changes over time. To retain essential information within the data while removing less informative elements to simplify and clarify the data, we will develop an automated deep learning model capable of establishing stringent data selection criteria. At the end, we will develop new techniques to make the deep learning more understandable to human. Through my research program, we anticipate several significant outcomes and benefits to the research field and to Canada as a whole. Firstly, I expect to develop advanced data analysis techniques that can combine different types of data into decision-making and will have broad applications across various domains, including healthcare, natural sciences, and engineering. Secondly, these techniques generate valuable data without privacy concerns, contributing to open data science and providing a widely shareable resource for other researchers. They also simplify complex data while retaining essential information, reducing the complexity of accessing and handling the data, which is beneficial for under-resourced communities. Thirdly, my research will make complex deep learning models more understandable. This will not only benefit the research community but will also make the results more transparent to stakeholders and decision-makers in Canada.