Quantitative and Computational Approaches in Translational Biophysics Applications

Abstract

Experimental and theoretical biophysical applications intertwine concepts from Computational, Life, Mathematical, and Physical sciences to expand the knowledge base and societal impact of various scientific advances. Interdisciplinary biophysics applications are generally unified by the underlying characteristics of the data they capture. Specifically, data created and collected in such applications have common characteristics of spanning space, time, and biological scales of life, rendering scientists with complex, high-dimensionality data to extract insights and draw conclusions. The harnessing of this data, development of widely applicable analytical tools, and transformation from raw data to tangible information is a challenge well suited for approaches based on statistical principles and advanced machine learning to tackle. This thesis covers the conceptualization, implementation, and application of quantitative approaches to analyzing biophysical data in biomedical contexts. Specifically, these projects span various biological scales and clinical levels, from pre-clinical data from ex-vivo samples and in-vivo animal models, through human patient data analyzed both post-hoc and approaching real-time. A primary focus of this work is the development and application of machine learning toolkits for predictive analysis, from data collection and engineering pipelines through model implementation, training, and analysis, using deep learning and model interpretability approaches to turn multidimensional biophysical data into actionable information. The main takeaway from this body of work is a set of new quantitative methods that outperform existing baselines for unique and novel, yet clinically relevant, biophysical datasets, unified by their translational potential and impact towards therapeutics and/or diagnostics