CHPC - Research Computing and Data Support for the University

Using Computation to Sort Through Billions of Single-Nucleotide Polymorphism Pairs Rapidly

By Randall Reese^1,2, Xiaotian Dai², and Guifang Fu²

¹Idaho National Laboratory, Department of Mathematics and Statistics, ²Utah State University

The interaction between two statistical features can play a pivotal role in contributing to the variation of the response, yet the computational feasibility of screening for interactions often acts as an insurmountable barrier in practice. We developed a new interaction screening procedure which is significantly more tractable computationally. Using the supercomputing resources of the CHPC, we were able to apply our method to a data set containing approximately 11 billion pairs of single nucleotide polymorphisms (SNPs). The goal of this analysis was to ascertain which pairs of SNPs were most strongly associated with the likelihood of a human female developing polycystic ovary syndrome. This distributed computing process allowed us to select in just over two days a set of around 10,000 SNP interaction pairs which factor most strongly into the response.  What may have previously taken several weeks (or months) to obtain now took less than three days. We concluded our analysis by using an implementation of multi-factor dimension reduction (MDR) on the previously aforementioned results.

System Status

Cluster Utilization