Abstract

In this work, we utilized network features of cancer gene interactomes to cluster pediatric sarcoma tumors and identify candidate therapeutic targets in an unsupervised manner. RNA-Seq data were mapped to protein-level interactomes to construct weighted networks for mathematical analysis. We employed a geometric approach centered on a discrete notion of curvature, which provides a measure of the functional association between genes in the context of their connectivity. Specifically, we adopted a recently proposed dynamic extension of graph curvature to extract features of the non-Euclidean, multiscale structure of genomic networks. We propose a hierarchical clustering approach to reveal preferential gene clustering according to their geometric cooperation which captured the characteristic EWSR1-FLI1 fusion in Ewing sarcoma. We also performed in silico edge perturbations to assess systemic response to simulated interventions quantified by changes in curvature. These results demonstrate that geometric network-based features can be useful for identifying non-trivial gene associations in an agnostic manner.