Outline of Dissertation. To overcome the disadvantages in all previous optical studies and better answer the mechanism for Co-C bond cleavage in the coenzyme B12-depedent enzymes, we developed a cryosolvent system to allow us to study the radical pair formation reactions in EAL from Salmonella typhimurium at low temperatures (234 K to 248 K) with real- time EPR spectroscopy. Chapter Two presents the development of the cryosolvent system, which includes three aspects as follows: (1) survey of cryosolvents; (2) application of the optimized cryosolvent to kinetically arrest the EAL·B12·substrate ternary complex; (3) instrumental setup for time-resolved, full spectrum continuous wave EPR in order to investigate the CoII-substrate radical pair formation reaction at subzero temperatures. After all the necessary preparations, we advance to Chapter Three, in which the temperature-dependence of substrate radical formation was examined from the original state (ternary complex) to the final state (CoII-substrate radical pair). In this chapter, more evidence for the formation of ternary complex after the sample preparation is presented. Another major focus in this chapter is searching for the mysterious 5’-deoxydenasyle radical state and building the kinetic model for the CoII-substrate radical pair formation based on the experimental result. This model is first, to our best knowledge, applied in the coenzyme B12-depend enzyme superfamily to obtain the thermodynamic parameters. Armed with time-resolved, full spectrum continuous wave EPR and 41% DMSO/water cryosolvent system and determined to obtain the microscopic rates for Co- C bond cleavage, we continued the hydrogen isotope effect study in Chapter Four. The experimental result demonstrates that the Co-C bond cleavage is the rate limiting step within the temperature range. A large positive entropy contribution in the Co-C bond cleavage is discovered, which provides insight into the mechanism of EAL. We also extrapolate the thermodynamics parameters to room temperature and show that the transition states of Co-C bond cleavage and HT1 possess similar free energy, which clarifies the role of HT1 in facilitating the Co-C bond cleavage under physiological conditions. Inspired by the X-ray crystallographic studies on diol dehydrase and Ramman spectroscopic studies on B12’s derivative, we desire to obtain the spectra of the ternary complex. The UV-visible optical method was chosen and a home-designed cryostat was made to couple in a Shimadzu 1600...
Outline of Dissertation. The outline of this dissertation is as follows: • In Chapter 2, we present the background information and literature review needed to understand and appreciate the work and contributions in the remainder of the dissertation. We introduce the core concepts related to machine learning before delving into machine learning applications in cybersecurity and, more specifically, ML-based malware detection. We then analyze the problem of adversarial machine learning and investigate how attacks against ML models are actually performed. Additionally, we review the defensive approaches that have been previously proposed for coping with this problem. We also provide a comprehensive literature review of related work to ours. • In Chapter 3, we propose StratDef, which is a novel strategic defense system for the malware detection domain based on a moving target defense approach. We overcome the challenges related to the systematic construction, selection, and strategic use of models to maximize adversarial robustness. We provide the first comprehensive evaluation of defenses against adversarial attacks on machine learning for malware detection, where our threat model explores different levels of threat, attacker knowledge, capabilities, and attack intensities. We show that XxxxxXxx performs better than other defenses even when facing the peak adversarial threat. We also show that, of the existing defenses, only a few adversarially-trained models provide substantially better protection than just using vanilla models but are still outperformed by StratDef. • In Chapter 4, for the first time, we study and compare the effectiveness of several recent MTDs against adversarial ML attacks. Under different threat models, we show that transferability and query attack strategies are still able to achieve high levels of evasion against these defenses. This is achieved by using a combination of prior attack strategies and novel ones that we propose in order to increase the evasion of models. We also show that fingerprinting and reconnaissance of MTDs is possible and demonstrate how attackers may obtain critical defense hyperparameters as well as information about how predictions are produced. Based on our findings, we present key recommendations for spurring future work on the development of effective MTDs for adversarial attacks in ML-based malware detection. • In Chapter 5, we present MalProtect, which is a novel stateful defense for the malware detection domain. We show that Mal...
