Bioinformatics: Improving the power of statistical tests for genetic interactions
This project aims to improve statistical tests for genetic interactions without main effects. These types of interactions, coined epistatic interactions are hypothesized to account for most of the phenotypic variation, and are therefore important to study.
While most contemporary methods only consider genetic interactions where at least one locus has a strong main effect, Jason H. Moore developed a method called Multi-factor Dimensionality Reduction (MDR) for interactions without main effects. It has been shown that MDR is more powerful than other common methods for discovering epistatic interactions.
We improve MDR by alleviating the need for a threshold parameter. This improvement give rise to a prediction-based test statistic that is directly related to the area under the ROC-curve, a common performance metric used in machine learning.
We will evaluate the test statistic on data from cardiovascular disease.