Network-Based Systems Pharmacogenomics: Methods and Applications

Large-scale molecular perturbational data provide signatures that represent changes on a cellular state from a systematic exposition to drugs or other forms of perturbations. Resources like the Library of Integrative Network-Based Cellular Signatures (LINCS) enable the identification of signature profiles important, e.g., for drug repositioning or target discovery based on automatic similarity searches across a vast reference profile space. In this thesis, we investigated the LINCS L1000 data repository consisting of nearly 2 million gene expression profiles and additional meta-data. As main results, we obtained: (I) an overview of the characteristics of all available data sets, their interrelations and experimental conditions including specific drugs, cell lines, time points and dosages. (II) a web interface called L1000 Viewer for accessing selected subsets of data from LINCS needed for the experimental design of studies addressing particular questions. (III) drug association networks (DANs) representing relationships for all drugs and small molecules in LINCS. The DANs are very informative in gaining a genomic-scale overview of the relationships among all drugs (including FDA approved drugs) and small molecules and, hence, provide a systems pharmacogenomic drug landscape. Importantly, we assessed the structural connectivity of the DANs by using information from the Anatomical Therapeutic Chemical (ATC) classification of drugs. This allowed us to identify the DAN modules’ as therapeutic attractors of ATC drug classes, extending the classic idea of cancer attractors in gene regulatory networks introduced by S. Kauffman to the compound space. In order to utilize our results, all DANs are available via an interactive web site allowing also the exploration of their structural complexity.