Statistical Methods for Analysis of High Throughput Experiments in Early Drug Development

Introduction: Advances in biotechnology and the ability to obtain molecular profiles of biological samples, and in particular, the transcriptomic data, have been transforming the way biomedical research and early drug development are carried out for more than a decade (Clarke et al., 2004; Chengalvala et al., 2007; Hughes et al., 2011). In view of increasing costs of the drug development and nevertheless a large number of drugs which fail the clinical trials either due to the lack of efficacy or side effects, novel technologies can potentially assist the selection of the most promising compounds for later stages in the drug discovery process. In particular, gene expression experiments are known to be of use for the compound target identification and side-effect profiling or toxicogenomics (Chengalvala et al., 2007). The target identification sheds a light on mechanism of action of a drug, whereas toxicogenomics can assist in prioritizing compounds for further development. The knowledge of gene expression activity induced by certain treatment can pinpoint genotypes which would either benefit from the treatment or have side effects in the late phases of clinical trials, thus, saving development costs. As pointed out by Xie et al. (2012); Ma and Zhao (2012a), cancer, neurological disorders and other complex diseases involve many genes and biological pathways. Therefore, development of multi-targeting therapeutics is needed for successful treatment of complex disorders. To obtain the knowledge of activated pathways, either on-target, desirable effects or off-target, possible side-effects, large numbers of gene expression experiments are carried out. In addition to the in-house data, i.e., data collected within a particular company, pharmaceutical industry may profit from the public data collections, such as Connectivity Map (Lamb, 2007) and Comparative Toxicogenomics Database (Davis et al., 2011). The efficient use of the data collections, either public or commercial, requires high quality data preprocessing and analysis tools. This PhD dissertation is dedicated to statistical aspects of data preprocessing and analysis. The dissertation consists of three parts. Part I focuses on the probe-level analysis of Affymetrix microarray data. Part II deals with biclustering analysis of gene expression data. Part III introduces joint biclustering methods for data integration and exploration.

Publication year:2013

Accessibility:Open