SUMMARY

Precision medicine has the potential to improve patient outcomes through customized clinical decisions for many diseases but is reliant on availability of robust biomarkers and assays for biomarker detection that can accurately quantify the disease state. Microfluidic devices are powerful diagnostic and research tools for functional testing of patient samples; these devices are increasingly sophisticated by incorporating physiological features such as the cellular environment, thus better recapitulating in vivo behavior. As these platforms continue to incorporate more features, analysis and interpretation of patterns within these multi-factorial datasets becomes challenging. For example, high-speed imaging technologies allows for the capture of high throughput quantitative data that incorporate dynamic signals and responses within a single experiment. Without interpretable models of these complex datasets, experimental observations are difficult to translate into clinically actionable insights for precision medicine. New statistical and computational methods are needed to extract the maximal amount of information from the analysis of microfluidics-generated data and overcome the challenges of modeling biological data. The overall objective of this thesis is to leverage computational and mathematical approaches to develop robust predictive models of patient sample response to combinatorial therapies assayed in microfluidic devices. I will validate my approach using microfluidics-generated datasets from application to two hematologic diseases: multi-drug resistance profiling in leukemia, and oxygen-dependent rheological biomarkers in sickle cell disease vaso-occlusion. The frameworks developed here will result in models that can extract important features from multi-factorial experiments, optimize discovery of synergistic interactions, and provide personalized recommendations for therapy.