Content area
Abstract
Precision oncology trials have demonstrated that matching cancer patients to treatment based on the molecular characteristics of their tumor leads to better response to therapy. Recent evidence suggests that transcriptomic profiling is particularly informative for matching patients to treatments. However, in many clinical studies, the majority of patients could not be matched to a genomically or transcriptomically selected therapy, highlighting a need for more biomarkers predictive of drug response.
Preclinical pharmacogenomic studies combine molecular profiling of cancer models with drug sensitivity screening with the goal of discovering novel predictive markers. However, inconsistencies between experiments done in different laboratories raised concerns about the reliability of biomarkers discovered from these datasets. In this thesis, I address three challenges to using such studies for the discovery of transcriptomic biomarkers predictive of drug response.
To address differences in data pre-processing and inconsistent nomenclature, I present Pharma-coDB, a unified database of preclinical pharmacogenomic studies. I then investigate which statistical approaches are best suited to correlating noisy drug response measurements, both in simulation and real datasets. I propose and investigate the rCI statistic, a semi-parametric extension of the concordance index designed to reduce the impact of noisy drug response measurements. The rCI is shown to be more powerful at detecting correlations than the concordance index or Spearman correlation coefficient, however, it remains less powered than the Pearson correlation in the application to drug response data. Finally, I present a framework using statistical meta-analysis models to resolve technical and biological differences across datasets. I apply this framework to the seven largest preclinical pharmacogenomic studies released to date, discovering 4338 putative robust trancriptomic biomarkers. Using clinical validation data, I demonstrate a novel biomarker, expression of ODC1, is predictive of paclitaxel response for treatment of Breast Carcinoma a neo-adjuvant setting.
Together, these contributions provide novel tools and propose novel approaches for analysing preclinical pharmacogenomic screens, as well as initiate the translation of discoveries made in these studies to clinically actionable biomarkers.
Details
