Title:	Robust disease modules
DNr:	LiU-compute-2020-21
Project Type:	LiU Compute
Principal Investigator:	Mika Gustafsson <mika.gustafsson@liu.se>
Affiliation:	Linköpings universitet
Duration:	2020-07-06 – 2022-09-01
Classification:	10203
Keywords:

Abstract

Background: an important aim of medical research is to find disease-specific genes for therapeutic and diagnostic purposes. In general, this is difficult because few genes are specific and the complexity of common diseases. This has motivated the study of disease modules by us and others, which basically assumes that diseases are due to perturbations of modules of highly interacting genes, which fulfills a specific molecular function (1). A key aspect of this is that the modules might be highly tissue specific, which has led our attention to focus our analysis to a tissue with great importance for many diseases, namely CD4 + T cells (2). Moreover, by analyzing disease module overlap of related diseases it might be possible to generate a better overview of the modules of interest. In previous work we have seen that the modules that overlap most are also the ones most enriched for genetic risk genes, biomarkers, and drug targets. However, the concept of disease modules is still not uniquely defined and different alternatives needs to be explored. Purpose and aims: This led us to hypothesize that combination of shared and specific modular components could be used to discriminate between a) patients and controls and b) different T-cell associated diseases. In this work we will analyze a prospectively collected data set of CD4+ T cells from about 180 patients and controls of 15 cell diseases. We will in this project we will explore alternatives for robust disease module characterization based data integration; how it should be defined; and how it varies across related diseases as asthma hay fever and eczema. We will test a) integration protein interaction network and inverse correlation methods, using the LASSO; b) clique based modules; c) seed based modules ; The aim is to define robust and biologically modules of the diseases using the power of similarities of the diseases, which then could be analyzed to define reliable entities which varies across diseases. Robust modules will be generated by using perturbations of the original data and re-calculating the procedure. Modules will then be defined in ways inspired by consensus clustering.