Graph classification is an important tool for analyzing data with structure dependency. In traditional graph classification, graphs are assumed to be independent where each graph represents an object. In a dynamic world, it is very often the case that the underlying object continuously evolves over time. The change of node attribute and/or network structure, with respect to the temporal order, presents a new structured sequential data representation. One of the major applications using machine learning algorithms (i.e., graph learning) is healthcare. Healthcare is a fast-growing domain for leveraging its massive structured and unstructured data. In this talk, I will first present some of my research work on large-scale graph and time series data, as well as how to analyze structured sequential data such as time-variant graph and networked time series. Next, I will introduce my applied research on healthcare, including autism screening, kidney function estimation and readmission prediction. The talk will be concluded with some of my ongoing projects and the potiential directions of future work.