Bilkent University
Department of Computer Engineering


Identification of Cancer Patient Subgroups via Pathway Based Multi-View Graph Kernel Clustering


Ali Burak Ünal
MS Student
Computer Engineering Department
Bilkent University

Characterizing patient genomic alterations through next-generation sequencing technologies opens up new possibilities for refining cancer subtypes. However, genomic alterations observed in cancer patients are highly heterogeneous which makes grouping patients based on solely on this information challenging. On the other hand, different genomic alterations may perturb the same functional mechanisms and different genomic data types provide complementary information. In this study, we propose a multi-view kernel clustering procedure that allows combining different types of genomic data in the context of multiple pathways through a novel graph kernel. In this framework, each biological pathway is represented as an undirected graph and for each patient the vertices of the graph are labeled based on her alterations. The similarities of patients are quantified through a smoothed shortest path graph kernel (smSPK), which is evaluated for each pair of patients by comparing their vertex-labeled pathway graphs. This results with multiple kernels derived from multiple pathways and genomic data types, which are then used as input to multi-view kernel clustering to cluster patients. Synthetic data experiments show that this approach is effective. Applying the methodology on ovarian cancer patients using mutational data and expression data reveals three clusters with patient groups that are significantly different in their survival times (p-value < 0.005).


DATE: 24 April, 2017, Monday @ 17:20