Computational Analysis of Transporter Repertoires as Determinants of Cellular Cancer Drug Response

Abstract

Cancer patient heterogeneities challenge disease management despite advances in targeted therapies. Patient sub-populations are irresponsive to certain treatments with unknown reason or differ in sensitivity, while mutations can cause resistance and patient relapse. Large-scale pharmacogenomic screens of large sets of small molecule libraries against comprehensive panels of human cancer cell lines were performed in order to provide novel predictive biomarkers of cancer drug response in the context of genomic profiles. However, predictive accuracy is still lower than desired. This master's thesis focuses on the roles of membrane transporters in cellular drug response. It has been hypothesised that many drugs act upon their endogenous targets by hitch-hiking on membrane channels. Solute Carriers (SLCs), which represent the second-largest family of membrane proteins in the human genome and the largest class of transporters, are the central focus of this study. About 10\% of the human genome encodes for transport-related functions, while a functional link between transporter and cargo and disease relevance are largely unknown. For systematic identification, the analyses also included more well-known transporter family of ATP-binding cassettes (ABC), which have a widely accepted role in mediating drug resistance. The landscape of expression for both transporter families in cancer tissues were analysed. Matrix factorisation based methods were also employed in integrating multiple genetic data and observe tissue specificity patterns. It was found that differential expression of several transporters are likely to link with tumorigenesis, Moreover, functions of transporters in drug influx and efflux were also computationally hypothesised. Several statistical methods were used and compared, with a list of most potential candidates suggested for experimental validation. The interaction between two transporters were also identified using linear model with regularisation with interaction terms. Both computational and biological challenges and limitations of the project are discussed.