Developing a data analysis pipeline for automated protein profiling in immunology

Abstract: Accurate information about protein content in the organism is instrumental for a better understanding of human biology and disease mechanisms. While the presence of certain types of proteins can be life-threatening, the abundance of others is an essential condition for an individual's overall well-being. Protein microarray is a technology that enables the quantification of thousands of proteins in hundreds of human samples in a parallel manner. In a series of studies involving protein microarrays, we have explored and implemented various data science methods for all-around analysing of these data. This analysis has enabled the identification and characterisation of proteins targeted by the autoimmune reaction in patients with the APS1 condition. We have also assessed the utility of applying machine learning methods alongside statistical tests in a study based on protein expression data to evaluate potential biomarkers for endometriosis. The keystone of this work is a web-tool PAWER. PAWER implements relevant computational methods, and provides a semi-automatic way to run the analysis of protein microarray data online in a drag-and-drop and click-and-play style. The source code of the tool is publicly available. The work that laid the foundation of this thesis has been instrumental for a number of subsequent studies of human disease and also inspired a contribution to refining standards for validation of machine learning methods in biology.