UC San Diego

Extrachromosomal DNA (ecDNA) is associated with aggressive tumor biology, but its role in tumor immune escape remains incompletely understood. We combined ecDNA annotations, RNA-seq data, somatic mutations, HLA presentation metrics, tumor purity estimates, and immune deconvolution results from a TCGA pan-cancer cohort. Our goal was to determine if ecDNA-positive tumors are immune-cold and to understand the mechanism behind this phenotype. Once adjusted for cancer type and tumor purity, tumors with ecDNA showed decreased cytolytic activity, a weaker CD8 signal, and reduced inferred cytotoxic immune infiltration, and these effects became more severe with higher ecDNA loads. Conversely, ecDNA-positive tumors did not reveal significant decreases in PHBR-based antigenicity metrics or strong binder fractions, which contradicts the notion of a broad loss of neoantigens. Although HLA-A and B2M showed selective decreases, the broader MHC-I transcriptional program was not coordinately suppressed. The most significant mechanistic finding was the sustained positive link between clonal MHC-II neoantigen load and overall cytolytic activity. However, this association was notably weakened in tumors containing ecDNA. These results indicate that ecDNA tumors are immune-cold due to a reduced ability to translate clonal MHC-II neoantigenicity into effective cytolytic immune activation, rather than a deficiency in antigenicity or widespread MHC loss.

The goal of predictive assembly is to use tunable components that can self-assemble into targeted structures. Components include, but are not limited to, solvent characteristics, ligand-particle parameters, and protocol design. Using computational methods, including DFT calculations, molecular dynamics, and machine learning, self-assembly components can be designed and in turn, structure can be predicted. The first chapter will address the solvent-aspect of self-assembly, where the structuring of a liquid can play a key role in self-assembling processes, such as protein folding and hydrophobic aggregation. Liquid structures can be quantified by the excess entropy, which is typically estimated using the 2-body entropy. We show that this method of estimating the excess entropy is insufficient in capturing the full structuring of liquid water, a common solvent, and using a modified form of the 2-Phase Thermodynamics Method, we can efficiently and accurately calculate the excess entropy of liquid water. In the second chapter, we use DFT calculations to predict binding energies of isocyanide ligands with different degrees of steric encumbrance. These DFT calculations are used to parameterize ligand-nanoparticle interactions and create a force field for MD simulations. These simulations capture the dynamics of ligands binding to nanoparticle surfaces. From these MD trajectories, we show how simulated Raman spectroscopy can grant insights into how well a ligand is bound to different sites on a nanoparticle. The last chapter addresses self-assembly at the mesoscale, where neuroevolutionary methods, like the L2G framework can be used to predict self-assembly protocols of tiling structures. This framework was restructured and adapted to predict protocols of exotic n-vertex tiling structures, which have been challenging to create experimentally. Additional fitness functions, such as a unitcell-based scoring metric, lead to improved protocol search for complex tilings. Lastly, we show how a pressure parameter was implemented to enable the L2G framework to better align with experimental parameters. Altogether, this body of work addresses aspects of self-assembly from the atomistic to mesoscale range.

This dissertation traces the history of pain research to tell a new origin story for the opioid crisis in North America. By charting the birth and afterlife of the most enduring theory of pain of the twentieth and twenty-first centuries, Ronald Melzack and Patrick Wall’s “gate control theory,” the dissertation argues that a holistic, post-Cartesian concept of pain made possible the overprescription of Purdue Pharma’s OxyContin and other opioids between the 1990s and 2010s. The dissertation begins with the scientific and political developments in modern Europe and North America that eventually destabilized a Cartesian concept of pain as a sensation of injury or disease. It subsequently shows how, over the mid-to late twentieth century, a new network of clinical and laboratory researchers concentrated in North America came to understand pain as an integrative experience of the entire nervous system, including a cognitive, emotional, and culturally enmeshed brain. Rather than just reduce the mind to the brain, these experts equated the material experience of pain with “the whole individual” and vice versa. Through this holistic concept of the body, a new international and interdisciplinary field of “pain research” advocated for long-marginalized patients, making their self-reports of pain, especially chronic pain, more visible than ever before. Yet, in its emphasis on the individual, this field also re-racialized pain. Turning away from an explicit language of “race,” its experts integrated “culture,” or “ethnicity,” into their holistic metaphysics to again mark bodies as categorically distinct and hierarchical in their experiences. By the 1990s, pain experts had mobilized their holism to pursue a campaign for expansive opioid prescribing for the chronic pains of implicitly white patients. Amid the war on drugs in the United States, a holistic metaphysics of pain carved out a new market for the pharmaceutical industry. In showing how pain became a personal experience, this dissertation explores what it calls neoliberal holism, or the “holistic” ideas, practices, and rhetoric through which the medical sciences have silently worked within neoliberal politics. In doing so, it considers alternative responses to the ongoing opioid crisis.

Scientific applications are increasingly using neural networks (NNs) at the edge as a fundamental tool for advancing fields such as particle physics and materials science. As the scientific instruments used in these experiments become more advanced, they produce a lot more data (e.g., 40 TB/s) than before. As a result, scientists are relying on edge NNs, which have more capabilities than traditional algorithms, to process the data. To process data quickly enough, these scientific edge NNS have unique requirements. They must (1) be heavily quantized and (2) execute fully on chip. Even more so, these scientific NNs often operate in high radiation environments (1000× that of space). My thesis focuses on using hardware-software co-design to create efficient, fault-tolerant computer architectures for NNs that execute fully on chip so that they meet the strict latency and throughput requirements laid out by these scientific experiments. I defend the following thesis statement: On-chip neural network inference introduces unique hardware-software codesign challenges and opportunities for building efficient, fault-tolerant neural network architectures. I provide evidence for my thesis in three parts: (1) codesigning residual NNs for efficient inference, (2) scaling up lookup-table NNs, and (3) codesigning fault-tolerant edge NNs. My thesis provides insights and tradeoffs that will help scientists better run their NNs on specialized hardware such as FPGAs and ASICs to advance research in their fields.

The sixth-generation (6G) cellular ecosystem is evolving toward integrated sensing and communication (ISAC), where the same infrastructure supports both high-rate data exchange and environmental perception. Vehicular networks are among the most demanding use cases: autonomous navigation demands ultra-reliable connectivity and centimeter-level positioning accuracy, yet conventional beam training protocols incur prohibitive overhead and global navigation satellite system (GNSS)-based localization fails in urban canyons. This dissertation develops a hybrid model/data-driven framework addressing these challenges through three contributions. The first introduces a passive radar-aided framework for multiuser millimeter wave (mmWave) link configuration. A roadside unit senses existing automotive frequency modulated continuous wave (FMCW) transmissions from multiple vehicles via a mixing filter bank with constant false alarm rate (CFAR) detection, isolating individual vehicle signals from multiuser interference. Three deep neural network architectures then map estimated radar spatial covariances to communication covariances, compensating for frequency and geometry mismatches between the 76 GHz radar and 73 GHz communication bands. The covariance-prediction-assisted approach reduces beam search space by up to 32 times and improves the achievable rate by 21.9% over radar-only methods. The second addresses high-accuracy 3D vehicle localization from a single base station snapshot. Two time-domain channel estimation algorithms, a two-stage multidimensional orthogonal matching pursuit (MOMP) variant and ESPRIT-D (an off-grid subspace method), extract multipath parameters while accounting for system filtering effects and unknown clock offsets. A lightweight network, PathNet, classifies estimated paths to select geometrically useful components, while a Transformer-based ChanFormer refines initial geometric position estimates through cross-attention, achieving 28 cm accuracy for 80% of users in line-of-sight (LOS) and sub-meter accuracy for 55% in non-line-of-sight (NLOS). The third extends single-snapshot localization to continuous tracking and collaborative multi-vehicle positioning. F-MOMP enables efficient high-resolution channel tracking by exploiting temporal correlation and factored dictionaries. VO-ChAT and VP-ChAT leverage channel sequence history through spatial-temporal and cross-attention mechanisms to track vehicle orientation and refine position estimates. For NLOS vehicles, a model-based collaborative framework exploits vehicle-to-vehicle (V2V) sidelinks to geometrically estimate per-vehicle clock and orientation offsets, achieving 0.3 m average error with sub-meter accuracy for 90% of cases. Together, these contributions demonstrate that embedding physical-layer signal models into deep learning architectures yields ISAC systems that are accurate, computationally efficient, and interpretable, key properties for safety-critical vehicular deployments.

In technology-mediated environments, perceptual information is filtered, transformed, or augmented, requiring observers to infer social, physical, and agentive states from perceptual evidence. I characterize this challenge as representation-reality alignment (RRA): The process through which humans coordinate perception, action, and inference to evaluate how perceptual representations correspond to underlying reality and to guide behavior accordingly. Across three chapters, I examine how children and adults interpret mediated information when perceptual input provides ambiguous or indirect evidence about the world.Chapter 1 asks how children aged 4 to 6 years infer a video chat partner’s visual access from the self-view. Children increasingly used the self-view as evidence of another person’s visual perspective with age; explicit reasoning emerged earlier than spontaneous action-based use. Chapter 2 asks how children aged 3 to 6 years judge whether perceptual features in augmented reality reflect physical reality or digital augmentation. Perceptual conflict increased active exploration, and exploration supported more accurate reality judgments; younger children relied more on physical inspection, whereas older children engaged more in causal testing to resolve perceptual discrepancies. Chapter 3 asks how adults infer internal capacities in a highly humanlike robot from cues of musical engagement. Holding the robot’s movements constant, musical-motion synchrony and contextual cues increased perceived warmth and competence of the robot, while reducing feelings of discomfort; effects were partly explained by beliefs about the robot’s ability to perceive and understand music.Together, these findings suggest that technology-mediated environments place common inferential demands on cognition, requiring observers to evaluate how perceptual information relates to underlying social, physical, and agentive states while coordinating perception, action, and inference to guide behavior. This dissertation advances RRA as a framework for understanding how people navigate technology-mediated environments.

Though activist groups in the San Diego-Tijuana area have long organized for Palestinian liberation, the continuous bombardment of Gaza by Israel and its resulting humanitarian crisis in the wake of October 7, 2023, has influenced a surge in transnational advocacy actions, oftentimes evoking comparisons or connections between the U.S.-Mexico border and borders surrounding Gaza and the West Bank. Analyzing events for Palestinian liberation and/or border advocacy between October 2023 and July 2024, this thesis reimagines the application of “linked fate” through a spatially removed, transnational lens motivated by interlocking systems and mechanisms of power. I propose that border sites offer tangible and theoretical links by which geographically disconnected communities may conclude that what happens in one impacts the other, clarifying how activists in the San Diego-Tijuana borderlands at times draw “linked fate” solidarity relationships through four common narratives: Directly, unpacking the material, financial, and law enforcement exchanges between the U.S. and Israel; Experientially, citing analogous experiences of life under settler colonialism, imperialism, and racial capitalism; Consequently, considering decisions at the state level which disparately impact communities abroad; and Interconnectedly, positing that liberation movements are inherently connected through intersecting systems of dominance.

The local-to-global principle—the idea that global structure is determined by, or can be certified from, local constraints—appears across mathematics, theoretical computer science, and physics under different disguises. This thesis studies two manifestations of this principle: in quantum error-correcting codes, where local parity checks certify the integrity of the global encoded information, and in the entanglement bootstrap, where local reduced density matrices determine the global structure of quantum phases of matter.Part I constructs quantum low-density parity check (qLDPC) codes. We first build a family of quantum locally testable codes (qLTCs) achieving linear dimension, near-linear distance, and near-constant soundness, by generalizing square-complex qLDPC codes to higher-dimensional cubical chain complexes. We then extend the cubical construction to the sheaf cochain complexes, on which we define Poincaré duality and a cup product. Poincaré duality clarifies the duality between the X- and Z-distances of the corresponding quantum code, while the cup product yields three sheaf codes that jointly support a transversal CCZ gate, providing a general route to non-Clifford gates on qLDPC codes.Part II develops an entanglement-based characterization of conformal field theories (CFTs). We first show that the ground state of any unitary 1+1D CFT is a critical point of a particular linear combination of subsystem entropies, and we verify numerically that this condition is stringent yet robust across known critical lattice models. We then leverage this characterization to perform a computational search over four-qubit and four-qutrit systems, which recovers known CFTs and identifies roughly 20 new candidate critical points, potentially including new irrational CFTs.

Genomic studies of high-altitude populations exposed to chronic low oxygen (hypoxia) have identified signals of selection linked to adaptive phenotypes. However, substantial variation both across and within populations suggests that human adaptation to hypoxia arises through multiple, distinct physiological and evolutionary pathways. Many Tibetans living at altitude exhibit lower hemoglobin and heightened ventilatory responses, whereas Andeans more commonly exhibit a higher average hemoglobin concentration and blunted ventilatory responses. Variation in hypoxia responses is shaped by factors such as the duration of high-altitude exposure (e.g., generations, years, days, hours), ancestry, and other biological factors and environmental influences. This dissertation examines human adaptation to altitude using a multi-omics framework designed to capture the heterogeneity of adaptive mechanisms. Chapter 1 examines convergent and divergent genomic signals of selection in Tibetan and Andean highland populations. Chapter 2 characterizes the transcriptomic landscape of individuals with chronic maladaptive excessive erythrocytosis undergoing isovolumic hemodilution. Chapter 3 investigates epigenetic remodeling during short-term acclimatization in unacclimatized lowland individuals across four days at altitude. Together, these studies identify previously unrecognized genotype-phenotype relationships and advance understanding of hypoxia biology with broad implications for human health, particularly cardiometabolic, pulmonary, and sleep physiology.

This dissertation develops a data-driven framework for discovering steady-state structures in parameterized dynamical systems. Motivated by applications in which governing equations may be unavailable, incomplete, or costly to solve repeatedly, we formulate equilibrium discovery as the problem of learning parameter-dependent structures directly from observational data. In particular, the relation between parameters and steady-state solutions may be multivalued and may change across parameter regimes. We introduce a scalar target function Φ(U,Θ) defined on the joint solution–parameter space. For each fixed parameter value Θ, this target function is constructed as a Gaussian-mixture landscape whose peaks encode the existence, multiplicity, and locations of steady-state solutions. To approximate this landscape, we develop the Parameter–Solution Neural Network (PSNN), an architecture designed to capture the joint but decomposable dependence on solution and parameter variables. We establish a universal approximation theorem with explicit error bounds for the PSNN, including target functions with different regularity in the solution and parameter directions. Building on the learned landscape, we develop a computational framework for recovering steady-state structure from the PSNN-predicted landscape. The framework includes a cutoff-and clustering-based algorithm for locating equilibrium, auxiliary classification models for predicting equilibrium counts and stability, and adaptive refinement strategies for resolving closely spaced steady states. These components allow the method to recover solution multiplicity, locations, and stability information even in challenging regimes where equilibria are close to one another or only partially observed. Numerical experiments on the Gray–Scott model and two-gene feedback loop systems demonstrate that the proposed methods accurately recover multiple steady states, phase boundaries, bifurcation behavior, and stability structure. The results further show that the framework remains robust under incomplete observational data, supporting its use as a data-driven approach for equilibrium discovery in nonlinear parameterized dynamical systems.