Application of Graph-based Data Mining to Metabolic Pathways

We present a method for finding biologically meaningful patterns on metabolic pathways using the SUBDUE graph-based relational learning system. A huge amount of biological data that has been generated by long-term research encourages us to move our focus to a systems-level understanding of bio-systems. A biological network, containing various biomolecules and their relationships, is a fundamental way to describe bio-systems. Multirelational data mining finds the relational patterns in both the entity attributes and relations in the data. A graph consisting of vertices and edges between these vertices is a natural data structure to represent biological networks. This paper presents a graph representation of metabolic pathways to contain all features, and describes the application of graph-based relational learning algorithms in both supervised and unsupervised scenarios. Supervised learning finds the unique substructures in a specific type of pathway, which help us understand better how pathways differ. Unsupervised learning shows hierarchical clusters that describe the common substructures in a specific type of pathway, which allow us to better understand the common features in pathways