This dissertation discusses the development of novel visualization techniques for biopolymer structures, the application of machine learning to biophysical data, and an in depth atomic-level investigation of a therapeutic nucleic acid called "F10."

The first part of the dissertation focuses on visualization and machine learning, addressing key issues of bias in the field of computational biophysics. The guiding principle in this work has been removing such bias and rigorously conveying uncertainty. To that end, the dissertation develops numerous methods for interpreting biopolymer ensemble data without the need for prior knowledge or setting of biasing parameters. These methods include grouping large sets of structure data into bins via clustering without the typical need for setting any biasing parameters; using a machine learning technique called "decision tree learning" to uncover the rules by which parts of proteins interact in a statistically rigorous, parameter-free, reproducible way; and creating simple visualizations of large data sets that tend to be too complicated for a human to easily understand using extant visualization methods.

For all of these projects, the dissertation provides a careful discussion of the limits of these methods and how researchers might visually convey the uncertainty inherent in them. Furthermore, it provides guidance on how to display what are effectively error bars on biopolymer structures. This part of the dissertation concludes with the development of a statistical tool to estimate the amount of sampling needed for any time-dependent multi-dimensional process without bias. These contributions may move the field forward in its ability to remove bias and convey uncertainty in statistically rigorous ways.

The second part of the dissertation focuses on understanding the effectiveness of and designing delivery mechanisms for therapeutic nucleic acids, particularly a 10mer of FdUMP (5-fluoro-2?-deoxyuridine-5?-O-monophosphate) -- also called F10. The machine learning techniques developed earlier in the dissertation are applied here to the ongoing drug discovery project that is F10. This therapeutic oligomer represents the culmination of over 60 years of rational drug design, starting with the development of 5-fluorouracil (5-FU), leading to the development of multiple delivery systems for this fluoropyrimidine, which paved the way for the development of F10. Compared to the currently used chemotherapeutic 5-FU, F10 is demonstrably more efficacious as a cytotoxic therapeutic, and better tolerated in vivo than the widely used 5-FU. That is, F10 is better at killing cancer but worse at killing people. This part of the dissertation begins with reproducing previous experimental results in silico, leading to a specific, detailed ordered of events for F10's zinc complexation. This work is of particular import, as zinc is also effective at killing cancer cells. Having reproduced -- and explained at an atomic level -- these experimental results, the dissertation moves to predicting F10's in vivo structural and kinetic properties in both intracellular and extracellular conditions. Then, the dissertation proposes a chemical perturbation of F10 that results in higher structural stability in the presence of certain salts. Finally, the work concludes with an investigation into a highly specific delivery mechanism for F10, which would increase its effectiveness at targeting cancer cells while further decreasing its effects on healthy cells, including suggestions for future investigations.