22689 research outputs found
Sort by
Understanding the Geologic Record at Gale Crater, Mars
No full textThe Martian sedimentary rock record provides a unique opportunity to study planetary environmental modifications through geologic time. Sedimentary rocks are formed when rock-forming particles (sediment) are lithified after being deposited through various surficial processes, making these rocks unique indicators of the environments that persisted at the time and place of deposition. The study of these layered packages of sedimentary rock (strata) is a main principle of geology known as stratigraphy. The sedimentary rock record of Mars is older than the terrestrial record, providing information about environmental conditions at the early stages of planetary evolution. Despite the clues left behind by rocks on the surface, many questions remain about the nature and evolution of Martian surface environments. Understanding the depositional history of Mars, and particularly its earliest environments as they pertain to the presence of liquid surface water and any associated habitability, are key questions pursued by the Mars science community. At the center of these ideas is Gale crater, where Mount Sharp, a 5 km high layered sedimentary package at the crater center, is actively being explored by the Mars Science Laboratory (MSL) Curiosity rover. Among Curiosity’s science payload is a robust imaging suite allowing for high-resolution reconstruction of stratigraphy through stereo-imagery. Utilizing these tools in addition to explorations of Martian analog sites on Earth, this thesis will aim to understand the geologic history at Gale crater. This will be addressed by pursuing the following questions: 1) What is the bedding orientation of layered stratigraphy at the base of Mount Sharp on the Vera Rubin ridge, 2) how are fine-grained, wind-blown sediments preserved in the rock record, and 3) what is the nature of stratigraphic transitions within Mount Sharp stratigraphy. Addressing these questions will aid in our understanding of the geologic history experienced at Gale crater, and our collective understanding of Martian environmental history
Exploring Limited Proteolysis Mass Spectrometry (LiP-MS) as a Tool to Study Protein Folding on the Proteome Scale
No full textProteins navigate intricate energy landscapes on their journey to attain their native structures. While past research focused on small proteins capable of reversibly refolding from a denatured state, it overlooked the complexity of natural proteomes, which consist of diverse proteins with complex structures. We seek to address which proteins can undergo reversible refolding from a denatured state and what the limits are for which thermodynamics can guide a protein’s folding. At the same time, we seek to provide an explanation for why refoldability might exist as a trait within proteomes.
We utilize Limited Proteolysis Mass Spectrometry (LiP-MS) as a tool to probe protein refolding kinetics for whole proteomes. This approach capitalizes on the fact that if a protein fails to refold, it will adopt a different structure than the native protein, resulting in distinct proteolysis patterns. Initial studies found that one-third of the E. coli proteome is not intrinsically refoldable. This brings into question whether exogenous factors and processes, such as chaperones or cotranslational folding are required for efficient protein folding. To answer this question, we developed a LiP-MS approach paired with an isotope-labeling strategy to globally monitor the structures of refolding E. coli proteins in the cytosolic medium and with the chaperones, GroEL/ES and DnaK/DnaJ/GrpE. Despite their different structures, these chaperones refold a similar set of proteins, suggesting they share a common mechanism for unfolding misfolded states. Additionally, some proteins remained resistant to refolding by either chaperone, suggesting that they may fold most efficiently cotranslationally and then remain kinetically trapped in their native conformations.
We further expanded the LiP-MS refoldability studies to two eukaryotic organisms (Saccharomyces cerevisiae and Neurospora crassa). In contrast to the bacterial proteome, these proteomes are more refoldable, which we hypothesize is due to the higher levels of intrinsically disordered regions (IDRs) within their proteomes. Finally, we demonstrate that proteins forming stress granules during heat-shock are generally more refoldable. with Hsp104 as a key mediator in disassembling these granules. Our data suggests that spontaneous refoldability is an adaptive trait that endows proteins with the capacity to reform their native soluble structures when extracted from condensates.
In summary, this thesis advances our understanding of protein folding, especially within diverse proteomes and highlights the role of IDRs in eukaryotic proteins. It provides insights into this essential biological process and its implications in native cellular environments
Delineating seizure networks using single pulse electrical stimulation
Get PDFEpilepsy is a complex neurological disease characterized by recurrent unprovoked seizures, affecting over 70 million people worldwide. Approximately one third of all epilepsy patients have drug resistant epilepsy, meaning their seizures cannot be controlled by medication alone. The most effective treatment for these patients is surgical removal, disconnection, or interruption of the epileptogenic zone, the brain region or regions responsible for generating the seizures. Therefore, accurately identifying the epileptogenic zone is essential to successful surgical treatment for these patients. The gold standard relies on passively observing spontaneous seizures while the patient is undergoing intracranial EEG monitoring, but this challenging process is costly, in terms of time and resources, and still results in limited rates of seizure freedom. Single pulse electrical stimulation provides an active method of probing the seizure network to map effective connectivity in the brain, and analysis of the responses elicited at local and distant sites may help identify excitable regions with high network influence. While this technique has the potential to supplement traditional passive intracranial EEG monitoring in delineating the epileptogenic zone, it is not widely adapted. This is due to the lack of a standardized methodology arising from an incomplete understanding of the effects of stimulation parameters, knowledge gaps in how to best analyze experimental results, and the need for validation of specific features as epileptogenic markers.
The research detailed in this dissertation demonstrates and further establishes the utility of single pulse electrical stimulation as a clinical tool for delineating seizure networks. This was accomplished first through a rigorous investigation of the effect of stimulation parameters on evoked responses, finding optimal current intensity, pulse width, and charge settings to maximize response amplitude and distribution. Next, the current-dependent excitability of the evoked responses was examined and used to discriminate epileptogenic sites. Then, network features of effective connectivity mapped by SPES were developed as markers of epileptogenic sites, using the mesial temporal region as an example region of interest. Finally, stimulation-induced spectral responses quantifying the excitability within epileptogenic networks were used to localize observed and targeted seizure onset zone sites and to predict surgical outcome
IN VIVO VOLTAGE IMAGING DURING BEHAVIOR COMBINED WITH TRANSCRIPTOMICS ENABLES DISSECTION OF INTERNEURON TYPES IN THE MOUSE MOTOR CORTEX
Get PDFThe aim of this work is to investigate the activity of inhibitory neurons in the anterior lateral motor cortex (ALM) of mice during a delayed response behavior and establish the relationship of activity patterns with transcriptomic types. ALM drives behavior in this task, and the function of ALM excitatory neurons is well-established. Here, we focus on two populations of inhibitory neurons, expressing the marker genes Ndnf and Vip, in cortical layers 1 and 2 of ALM, for which no previous measurements of in vivo activity have been made.
To achieve this goal, we developed in vivo voltage imaging in behaving animals, which offers better experimental throughput for measuring in vivo activity of sparse, superficial neurons compared to existing methods. We used Voltron, a chemigenetic voltage indicator, and confirmed that it reports spikes and subthreshold membrane potential in vivo. We validated the accuracy of voltage imaging signals in vivo with simultaneous whole-cell patch clamp measurements. We built an analysis pipeline for estimating spike times and membrane potential from voltage imaging data, which enhances our ability as a field to perform in vivo voltage imaging and interpret the results.
We found that Ndnf+ and Vip+ neurons exhibit task-modulated activity. Ndnf+ neurons can be classified into subgroups with positive activity correlations within each subgroup and negative correlations across subgroups. We found sparse connectivity between Ndnf+ neurons, consistent with their correlated activity patterns.
We identified types of inhibitory neurons using post hoc mRNA in situ hybridization to measure gene expression, and we are currently investigating whether different transcriptomically defined types of neurons show varying activity patterns during behavior. While we have not yet obtained activity measurements and reliable transcriptomics data in the same neurons, work towards this goal is ongoing
Manipulating Emotions: Generative Modeling of Prosody for Emotional Speech Synthesis
No full textThis thesis focuses on data-driven emotional speech generation using prosodic elements. Speech, a universal mode of communication, carries vital information beyond semantics, such as speaker identity and emotion. This work emphasizes intonation, intensity variation, and rhythm modulation as key prosodic elements for emotional speech understanding and generation.
Throughout this thesis, we combine probabilistic modeling with deep neural networks to transform neutral speech into emotional speech. We explore supervised, unsupervised, and reinforcement learning paradigms and rigorously evaluate these techniques against state-of-the-art models. The VESUS corpus, a reference dataset collected by us, ensures our methods generalize across multiple speakers and unseen vocabulary.
Chapter 1 introduces speech production, emotion models, and the importance of prosody in emotional speech perception. It sets the stage for emotional speech generation using prosodic transformation.
Chapter 2 provides essential technical background on diffeomorphic mapping, variational inference, and graphical models. These concepts are crucial for understanding the subsequent chapters.
Chapter 3 and Chapter 4 focus on supervised models for modifying prosody (F0 and energy) using the VESUS corpus. The former presents two models: one based on a highway network with gender embeddings and another employing diffeomorphic regularization. Chapter 4 extends the model to predict F0 and energy contour at the utterance level, leveraging segmental and supra-segmental properties.
Chapter 5 introduces the Variational CycleGAN framework for unsupervised prosody modification, addressing the limitations of vanilla CycleGAN.
Chapter 6 presents a supervised rhythm modulation algorithm combining generative modeling and dynamic time warping (DTW) to align input speech with a hypothetical desired output. It uses latent variable modeling for attention maps and DTW similarity matrices.
Finally, Chapter 7 discusses an unsupervised duration modification method employing reinforcement learning. This approach identifies important segments within an utterance using a masking strategy with a first-order Markov property, with the agent learning a distribution over transformation options.
In summary, this thesis employs diverse techniques, from supervised to unsupervised learning, to enhance emotional speech using prosodic elements, culminating in comprehensive evaluation and applicability to various speakers and vocabulary
Exploiting Structural Properties in the Analysis of High-dimensional Dynamical Systems
Get PDFThe physical and cyber domains with which we interact are filled with high-dimensional dynamical systems. In machine learning, for instance, the evolution of overparametrized neural networks can be seen as a dynamical system. In networked systems, numerous agents or nodes dynamically interact with each other. A deep understanding of these systems can enable us to predict their behavior, identify potential pitfalls, and devise effective solutions for optimal outcomes. In this dissertation, we will discuss two classes of high-dimensional dynamical systems with specific structural properties that aid in understanding their dynamic behavior.
In the first scenario, we consider the training dynamics of multi-layer neural networks. The high dimensionality comes from overparametrization: a typical network has a large depth and hidden layer width. We are interested in the following question regarding convergence: Do network weights converge to an equilibrium point corresponding to a global minimum of our training loss, and how fast is the convergence rate? The key to those questions is the symmetry of the weights, a critical property induced by the multi-layer architecture. Such symmetry leads to a set of time-invariant quantities, called weight imbalance, that restrict the training trajectory to a low-dimensional manifold defined by the weight initialization. A tailored convergence analysis is developed over this low-dimensional manifold, showing improved rate bounds for several multi-layer network models studied in the literature, leading to novel characterizations of the effect of weight imbalance on the convergence rate.
In the second scenario, we consider large-scale networked systems with multiple weakly-connected groups. Such a multi-cluster structure leads to a time-scale separation between the fast intra-group interaction due to high intra-group connectivity, and the slow inter-group oscillation, due to the weak inter-group connection. We develop a novel frequency-domain network coherence analysis that captures both the coherent behavior within each group, and the dynamical interaction between groups, leading to a structure-preserving model-reduction methodology for large-scale dynamic networks with multiple clusters under general node dynamics assumptions
Teleoperation Methods for High-Risk, High-Latency Environments
Get PDFIn-Space Servicing, Assembly, and Manufacturing (ISAM) can enable larger-scale and longer-lived infrastructure projects in space, with interest ranging from commercial entities to the US government. Servicing, in particular, has the potential to vastly increase the usable lifetimes of satellites. However, the vast majority of spacecraft on low Earth orbit today were not designed to be serviced on-orbit. As such, several of the manipulations during servicing cannot easily be automated and instead require ground-based teleoperation.
Ground-based teleoperation of on-orbit robots brings its own challenges of high latency communications, with telemetry delays of several seconds, and difficulties in visualizing the remote environment due to limited camera views. We explore teleoperation methods to alleviate these difficulties, increase task success, and reduce operator load.
First, we investigate a model-based teleoperation interface intended to provide the benefits of direct teleoperation even in the presence of time delay. We evaluate the model-based teleoperation method using professional robot operators, then use feedback from that study to inform the design of a visual planning tool for this task, Interactive Planning and Supervised Execution (IPSE). We describe and evaluate the IPSE system and two interfaces, one 2D using a traditional mouse and keyboard and one 3D using an Intuitive Surgical da Vinci master console. We then describe and evaluate an alternative 3D interface using a Meta Quest head-mounted display. Finally, we describe an extension of IPSE to allow human-in-the-loop planning for a redundant robot. Overall, we find that IPSE improves task success rate and decreases operator workload compared to a conventional teleoperation interface
Mitochondrial Homeostasis and Cellular Stress Resistance
Get PDFMitochondria contain a dedicated genome (mtDNA) comprising several genes necessary for respiration, mitochondrial transcription and translation, and other vital functions. Various stressors can destabilize mtDNA leading to mtDNA loss. While some cells can survive mtDNA loss, they exhibit various deficiencies. Here, we investigated the impact of proteotoxicity on mitochondrial function by inducing mitochondrial unfolded protein stress in budding yeast. This led to rapid mtDNA loss, but aerobic conditioning imparted transient resistance to mitochondrial protein stress. We present a quantitative model of mtDNA loss in a growing cell population and measure its parameters. To identify genetic adaptations to mtDNA depletion, we performed a genome-wide screen for genes for which a dosage increase affects the growth of cells lacking mtDNA. The screen revealed a set of dosage suppressors that alleviate the growth impairment in mtDNA-deficient cells. Additionally, we show that these suppressors of mtDNA stress both bolster cell proliferation and prevent mtDNA loss during mitochondrial protein stress. Follow-up studies explore a functional genomics approach to engineering resistance to other cellular stresses and discuss the relevance of stress resistance to longevity, using budding yeast as a proof of concept
MU OPIOID RECEPTOR AGONISM ELICITS EXCITATION IN ITCH SELECTIVE SENSORY NEURONS
Get PDFItch is a fundamental sensory signal for most animals, as an external extension of the
immune system, to rid the body of harmful organisms. The study of itch has been growing in
the past few decades, but the ligands and receptors responsible for it have yet to be completely
identified and characterized. This is especially true in the clinical setting, where chronic itch
conditions often lead to a big collection of itch-associated chemicals that build up in the skin
and together induce unwanted itchy side effects. One such compound is the endogenous
peptide, Beta Endorphin, which is known to bind the μ-Opioid-Receptor. Beta Endorphin’s role
as a pruritogen, or itch inducing substance, has not been fully investigated as it relates to
sensory nerves; nor has the role of peripherally expressed μOR been elaborated in the context
of itch. In this dissertation, I will first provide a general overview of the neural mechanisms of
itch. Following that, I will discuss a key method, intravital calcium imaging, used in my
experiments to investigate itch. I will then provide a more focused overview of the known roles
of opioids in the context of itch. The dissertation will then include my own results from
experiments that provide evidence for Beta Endorphin’s role in inducing itch via the μOR,
especially as it relates in a mouse model of psoriasis
Do rules rule tense learning in people with Williams syndrome? Tense knowledge in typical and atypical development
Get PDFIn this dissertation, I ask the question “What are the cognitive mechanisms involved in learning the tense system in English, both in individuals with Williams syndrome as well as typically developing individuals?” Williams syndrome is particularly interesting in this domain due to the complexity of the linguistic profile in the syndrome, especially regarding claims that language is spared in light of impaired spatial cognition. I first review two major theories of tense learning, processing, and production—the Words & Rules theory and Albright and Hayes’ minimal generalization model—as well as briefly discuss Rumelhart and McClelland’s connectionist model. I then discuss the contributions of memory to tense learning and processing. Next, I present three novel experiments investigating the relationship
between memory and tense in people with Williams syndrome and typically
developing children. These experiments contribute significantly to our understanding of the cognitive processes involved in learning and producing the past tense. My results suggest that (1) a dual mechanism theory of tense is not a sufficient account of the tense system in WS nor in typical development, and (2) the underlying tense system in individuals with Williams syndrome is remarkably similar to typically developing individuals