With the development of omics technologies, there is a large amount of biological data available. The biological data is usually complex, ill-sampled, and high dimensional. As a result, gaining insightful knowledge from the biological data is still a challenging problem. Many of those biological data can be represented using graphs. Over the years, many graph-based machine learning methods have been developed to analyse graphs in many tasks, such as module detection, feature engineering, and link prediction. This thesis provides the applications of graph-based machine learning methods on three types of biological data with the following contributions:
• A novel method that allows overlapping between the peak modules
in the metabolite annotation problem on liquid chromatography-mass
spectrometry untargeted data.
• A novel method that allows high freedom of the shapes of the topological modules in the detection of network biomarkers for gene expression
data.
• A novel method that incorporates the implicit networks constructed
from the drug-target interaction network in the drug-target interaction
prediction problem.
With these contributions, we can gain more understanding of the biological
data.