Abstracts Category : Other

Quantitative modeling and analysis of drug screening data for personalized cancer medicine

by Bhagwan Yadav

Despite recent progress in the field of molecular medicine, the treatment and cure of complex diseases such as cancer remains a challenge. Development of resistance to first-line chemotherapy is a common cause of current anticancer treatment failure. To deal with this problem, the personalized medicine (PM) approach has been adapted toward more targeted cancer research and management. The PM approach is based on each patient s genetic, epigenetic and drug response profiling, which is used to design the best treatment option for the given patient. As the PM approach is increasingly being adopted in clinical practice, there is an urgent need for computational models and data mining methods that allow fast processing and analysis of the massive relevant profiling datasets. High-throughput drug screening enables systematic profiling of cellular responses to a wide collection of oncology compounds and their combinations, hence providing an unbiased strategy for personalized drug treatment selection. However, screening experiments with patient-derived cell samples often results in high-dimensional data matrices, with inherent sources of noise. This complicates many downstream analyses, such as the detection of differential drug activity or understanding the mechanisms behind drug sensitivity and resistance in a given patient. To meet these challenges, a computational pipeline for drug response profiling was developed in this thesis. The pipeline was based on a novel metric to quantify drug response, called the drug sensitivity score (DSS). Further, by combining the normalized drug response profile of each cancer sample with a global drug-target interaction network, a target addiction score (TAS) was developed to de-convolute the selective protein targets and obtain knowledge on their functional importance. Finally, delta scoring was developed to quantify drug combination effects and to address the problem of the clonal evolution of cancer, which often leads to resistance to mono therapies. This novel computational pipeline improves understanding of cancer development and translates compound activities into informed treatment choices for clinicians. As exemplified in two case studies of adult acute myeloid leukemia (AML) and adult granulosa cell tumor (AGCT), the models developed here have the potential to significantly contribute to the effective analysis of data from individual cancer patients and from pan-cancer cell line panels. Hence, these models will play a substantial role in future personalized cancer treatment strategies and the selection of effective treatment options for individual cancer patients. NA