Finding Global Contrast Core Subgraphs in Large-Scale Genetic Association Study

Abstract

Genetic association study (GAS) is crucial to reveal the underlying principles of complex diseases. This is the first work that introduces contrast subgraph mining in two networks with significantly different edges or edge weights but the same vertices to solve GAS problem. It can not only detect the genetic loci highly associated with certain diseases, but discriminate between disease causing and disease preventing genetic loci, which captures more comprehensive and informative value for biologists. Inspired by the concept of r-core and minimum vertex weight, we propose to identify the novel global contrast r-core subgraphs r-GCCSs, which is more robust to outliers and redundancy. Further, we formulate the top-k r-GCCSs detection problem based on global contrast measure. In particular, (1) a linear time search algorithm is carefully developed to find the top-k r-GCCSs; (2) To further reduce the high computational cost, a linear space index is devised to support the top-k search. Comprehensive experiments on four large-scale real datasets demonstrate the efficiency and effectiveness of our approaches.