An MDL analysis framework for eQTL data

Rapid development of genome sequencing technologies enables novel insights into the mechanisms of complex disease through Big Data analysis. Physicians can nowadays assay a patient's gene variants and gene expression patterns in a timely manner and use the obtained data to study an individual's susceptibility to complex disease and unravel the underlying mechanisms of disease pathogenesis. Massive amounts of correlated genotype, gene expression, and clinical data are collected in eQTL datasets. In this work, we propose an analysis framework based on the minimum description length principle for extracting useful information from eQTL data. This is achieved by minimizing the stochastic complexity of the data by using the universal normalized maximum likelihood code as the global code length optimization criterion. The algorithm simultaneously identifies disease associated features, extracts the optimal model of the complex disease, and estimates its parameters. Applied to a simulated eQTL dataset, our framework successfully reveals the underlying mechanisms of a hypothetical complex disease interaction network.