Computational approaches for biological data integration

Biological processes in a cell are highly dynamic and their regulation involves a multitude of molecular components such as DNA, genes, proteins, and metabolites. It is of critical importance to understand these entities not only as separate elements but also in terms of their interactions with one another. Technological developments have enabled the rapid generation of vast volumes of data for nearly all types of such entities from an individual, referred to as ‘omics’ data. This thesis addresses three main challenges for the integrative analysis of different omics data modalities while taking their inherent heterogeneity into account: (i) ease of accessibility of high-throughput data so that it can be combined with in-house experimental data or used for reanalysis, (ii) integration of information across different resources, and (iii) integration within or across data modalities using networks. Chapter 2 of this thesis describes our approach to build a compendium of functional genomics data retrieved from GEO. With the associated R package compendiumdb and the accompanying MySQL database, pre-processed GEO data from different studies and profiling platforms can be systematically retrieved and stored. Chapter 3 of this thesis describes a problem-driven integrative analysis approach across different data sources to rank candidate proteins for low-abundant spots in 2D-DIGE experiments. Chapter 4 of this thesis describes a network-based integration method to align a pair of gene coexpression networks generated from gene expression data measured across multiple conditions. The method is applied to gene expression data measured in human and mouse immune cell types to study conservation and divergence between the two species. In Chapter 5 of this thesis a special case of this method is used to identify modules conserved between species for a single condition. The method is applied to gene expression data measured in human and mouse livers.