Title: Detecting outliers on microarrays with mixtures of normal and heavy-tailed distributions
Authors: Alexandra Posekany - University of Technology Vienna (Austria) [presenting]
Abstract: A common assumption in statistical modelling is a normal error distribution, however many data sets in fields like biology or economics do not follow these, independent of sample sizes. To robustify inference we employ a Bayesian hierarchical mixture model which simultaneously performs bio-informatical inference and detects outliers on a gene and array level. This is obtained by mixing normal mixture components with Student's $t$ distributed ones to identify the over-dispersed part of data which may originate from biological processes or laboratory work. In our application, we present several microarray data, which generally show over-dispersed noise behaviour. In recent years, microarrays were less regarded in molecular biological research, but have been introduced to clinical practice which makes the detection of outliers indicating problems in the medical and bio-informatical analyses even more relevant. In addition to provide a better inference of differential expression, the goal is to identify noisy genes in a gene and array level. Thus, we wish to identify whether single arrays are responsible for this behaviour to provide a quality control for clinical practice.