The food industry is one of the largest in Sweden and through it, large side streams are being generated. One such is bran, derived from cereal grains, composed of extractable dietary fibres capable of becoming novel materials such as laccase-driven crosslinked hydrogels. This thesis will explore the valorisation of these fibres through enzymatic modifications, as novel delivery vessel and potential prebiotic properties in order to integrate significant food losses into a circular economy with added health benefits towards the populace.

Feruloylated arabinoxylans previously extracted from wheat (WAX) and rye (RAX)were treated with selective arabinofuranosidases to successfully remove arabinose substitutions not containing ferulic acid moieties, exposing the xylan backbone. This lead largely to a reduced degree of crosslinking in WAX and RAX hydrogels. Rheological and morphological features were affected by the removal with an increase in viscoelastic properties in RAX hydrogels and lead to a more ordered network structure while WAX hydrogels saw the opposite effect regarding both features.

Residual β-glucans in WAX and RAX extracts were investigated. Their molecular weight differed from commercially available β-glucans derived from cereal endosperm, but ratios of cellotriosyl and cellotetraosyl remained similar to previously reported β-glucans. Removal of β-glucans from WAX and RAX extracts using a lichenase prior to laccase-crosslinking resulted in lowered viscoelastic properties for WAX hydrogels and an increase for RAX hydrogels. The removal likely exposes the arabinoxylan backbone similarly to arabinose removal. Freeze-drying at lower temperatures prior to regeneration of hydrogels lead to decreased pore size in hydrogels but increased porosity in RAX hydrogels and decreased in WAX hydrogels,which in turn lead to a decrease in viscoelastic properties in WAX hydrogels but increased in RAX hydrogels. Difference in chemical and physical interactions between the arabinoxylan network between WAX and RAX is believed to be behind their different behaviours. WAX and RAX hydrogels successfully encapsulated and retained target biomolecules of varying sizes and properties as a proof of concept. Pore size and porosity did not severely affect retention of smaller molecules such asglucose and tryptophan but for larger proteins. Addition of a gut bacteria xylanase triggered release of encapsulated molecules.

The common gut bacteria Bacteroides ovatus was successfully grown on WAX, RAX and CAX (corn arabinoxylans) extracts and hydrogels, producing health beneficial short chain fatty acids as secondary metabolites, highlighting the potential prebiotic properties of the extracts and hydrogels. Structural and biochemical differences between carbon sources affected growth, metabolite production and expression levels for enzymes related to arabinoxylan degradation. This thesis has demonstrated that dietary fibres can be effectively valorised by tuning them for multifunctional hydrogels contributing to a circular economy and improved public health.

Proteins are diverse biological macromolecules that play essential roles in many biological functions. The study of proteins has led to important biological and medical discoveries. With the advancement of technology, an increasing number of proteins can be measured in a single experiment. Today, from just a drop of blood, we can measure hundreds or thousands of proteins in the study of proteomes. This field of study is called proteomics and allows us to survey the vast array of proteins in our bodies to discover biological relationships that can improve our understanding of health and disease.

As the number of proteins that we can measure increases, so does the burden of analysing the increasingly large data sets. Data analysis pipelines must be crafted with care at every step, from data preprocessing to analysis, visualisation, and presentation. This thesis presents studies that showcase how one may go about interacting with large affinity proteomics data sets. 

To study the proteome we must be able to reliably measure proteins. Antibodies are used heavily in affinity proteomics for their excellent sensitivity. Their selectivity, however, must be validated thoroughly to ensure that we are measuring the correct protein. In study A, we validate antibodies targeting a clinically important family of proteins. The results have been published in an interactive web application open for anyone to browse.

The type of biological sample we use influences what proteins are present and what research questions we can answer. Blood is a practical sample type for its minimal invasiveness and its ability to provide a systemic view of an individual’s health. Dried blood spots (DBS) can be used as an alternative to venous blood draws that may also be performed without medical expertise, allowing remote self-sampling. In study B we perform a population study during the COVID-19 pandemic using DBS and demonstrate the feasibility of the sampling method for population proteomics. Study C expands on the previous study, establishing a general pipeline for immune phenotype exploration. These studies demonstrate the utility of the sampling and data analysis methods for profiling immune phenotypes through serology and proteomics. In study D we explore the biological differences between DBS and blood plasma in a large proteome survey using multiple proteomics technologies. Together, these studies provide insights into the dried blood proteome and how such data may be processed and analysed.

While the full breadth of proteomics and the analysis of such data cannot be captured in this one thesis, the work presented herein provides important contributions toward the reproducible analysis of large-scale affinity proteomics data.

Increasing cellulose accessibility and reactivity can expand its use beyond paper products, facilitating the development of high‑performance derivatives and regenerated cellulose, as well as efficient pathways for degrading cellulose to glucose. Structurally, cellulose consists of β‑1,4‑linked D‑glucopyranoside units, each of which contains three available hydroxyl groups. However, these hydroxyl groups are not fully accessible due to cellulose’s crystalline structure. This limits reactivity, which is a key parameter not only for derivatization but also for regeneration and efficient degradation. Swelling or partial dissolution of cellulose in sodium hydroxide followed by reprecipitation has been shown to disrupt the ordered structure and increase the exposure of reactive sites. Yet the swollen material retains large amounts of water, and upon drying, it is susceptible to hornification, which reduces its ability to reswell and lowers its reactivity. This thesis explores a cold‑alkali swelling method designed to increase the reactivity across a wide range of feedstocks. A proof-of-concept with microcrystalline cellulose established strategies to mitigate hornification while preserving the swollen structure after drying. The method was then applied to paper‑grade pulps and recycled textile waste, demonstrating that these lower‑grade materials can be upcycled into more reactive cellulose suitable for derivatization, regeneration, and cellulose degradation.

This thesis presents a multi-platform investigation of auroral processes, combining in situ measurements, satellite-based optical observations, and satellite-receiver signal analysis.

First, data collected by the SPIDER-2 sounding rocket and its deployed subsystems are analysed to examine the properties of pulsating aurora, providing detailed insight into its altitude, densities, pulsating frequencies and modulation. Combined measurements of Langmuir probes and a wave propagation experiment reveal the altitude profile of the electron density, revealing the peak electron densities of the order of 10¹¹ m^-3 at about 100 km altitude. An ion chemistry model allows to derive the energy profile of the precipitating electrons, which is found to peak at about 19 keV. In addition, ion probe and photometer measurements resolve individual pulsations with characteristic periods of about 2 s.

Second, limb observations from the Swedish MATS satellite are used to conduct a statistical study of a particularly poorly understood auroral emission, the O₂ atmospheric band at 762 nm. The results reveal a dependence between the geomagnetic activity and magnetic latitude, between the peak emission altitude and the magnetic local time, and between the intensity of the emissions and the emission altitude, providing new information on the characteristics of this auroral emission across both hemispheres.

Third, disturbances in Global Navigation Satellite System (GNSS) signals are shown to correlate with intensity variations in pulsating aurora, and a novel large-scale statistical analysis of this phenomenon is performed. The study, based on 14 years of data from Swedish receiver stations at auroral latitudes, demonstrates that total electron content (TEC) variations with periods of 33 s and shorter exhibit dependencies consistent with auroral activity in terms of solar cycle, seasonal, and diurnal behaviour. By identifying events associated with pulsating aurora, new relationships are established, including the dependence of pulsation period on magnetic local time, geomagnetic activity, and solar wind conditions, as well as a scaling between TEC fluctuation power and geomagnetic activity.

In summary, this work presents three complementary approaches to studying aurora: a sounding rocket case study, a statistical analysis from satellite limb observations, and an investigation based on GNSS signal disturbances. Together, they bridge small-scale in situ measurements and large-scale observations, advancing the understanding of pulsating aurora and the O₂ auroral emission.

Uranium mononitride (UN) is a promising advanced nuclear fuel for Gen-IV fast-spectrum systems and small modular reactors due to its high uranium density and superior thermal conductivity compared to conventional oxide fuels. However, its behaviour under irradiation, particularly the coupled effects of fission product (FP) chemistry and radiation damage on thermophysical and mechanical properties, remains insufficiently understood.

This thesis presents an experimental investigation of the thermophysical, mechanical, and dimensional response of pure and doped UN within a separate-effects testing framework. Simulated burn-up fuels (SIMFUEL) were employed to study the role of chemical disorder, while ion irradiation and implantation were used to introduce controlled radiation damage and selected FP chemistry.

Chemical modification was achieved through powder mixing and arc-melting routes, enabling the incorporation of zirconium and thorium as solutes. Structural characterization confirmed the formation of solid solutions, and secondary phases as well as microstructural variations depending on processing route. Thermal diffusivity measurements revealed that chemical disorder leads to a systematic reduction in thermal transport, primarily attributed to enhanced phonon scattering resulting from mass and strain fluctuations in the lattice.

Irradiation effects were investigated using proton irradiation and ion implantation of selected FP (Zr, Ba, Kr, Xe). Defect formation, clustering, and irradiation-induced cracking were observed, with clear dependence on damage level and implanted species. Near-surface thermal diffusivity measurements demonstrated a degradation of heat transport with increasing irradiation dose. Mechanical characterization revealed irradiation-induced hardening, while microstructural observations showed crack initiation and propagation associated with defect accumulation and stress development.

In addition, a surface-based methodology was developed to quantify irradiation-induced dimensional changes at the microscale. Both swelling and shrinkage were observed, highlighting the complex relation between defect evolution and microstructural conditions.

The combined results establish clear relationships between chemical disorder, irradiation damage, and property degradation in UN. The findings demonstrate that both compositional modification and defect accumulation significantly influence thermal transport, mechanical response, and dimensional stability. 

This work provides new experimental insight into the behaviour of UN and contributes to the understanding required for its future application in advanced nuclear energy systems. The results motivate further studies combining chemistry and irradiation with in-reactor validation to fully capture fuel performance under realistic reactor conditions.

This doctoral thesis presents a comprehensive summary of research efforts with the aim of advancing the state-of-the-art in decentralized optimization. As modern distributed systems grow in scale and complexity, traditional optimization methods face significant bottlenecks. This work, presented as a compilation of five papers, specifically targets the Alternating Direction Method of Multipliers (ADMM). The central objective is to re-engineer ADMM to overcome the dual challenges of convergence latency and communication inefficiencies in peer-to-peer networks.  The first part of the thesis focuses on decentralization, convergence speed, and computational cost. While centralized ADMM algorithms struggle with scalability and bottlenecks, existing decentralized schemes rely on either excessive computations and messaging or completely sequential operations, causing a prolonged time required to reach convergence. To tackle these problems, we introduce two fast-converging decentralized ADMM algorithms. Our theoretical analysis confirms that our algorithms retain the classical convergence properties of centralized ADMM while maintaining a low per-node complexity of $O(1)$. Numerical simulations further demonstrate that our algorithms converge significantly faster than state-of-the-art decentralized implementations, providing clear conditions under which they outperform traditional benchmarks.  The second part of the thesis addresses the straggler problem, which arises from the conventional ADMM requirement for global synchronization, where faster nodes are forced to remain idle until the slowest nodes complete their local updates, impeding the progress of the entire network. Here, we introduce three algorithms to allow the system to remain productive even under single-point-of-failure scenarios or extreme hardware variance. The first algorithm achieves straggler-resilience by allowing the nodes to proceed to the next iteration even when one or more nodes have not provided an update for one or more iterations. The second algorithm is a decentralized version of the first algorithm. The second algorithm enforces fast convergence as well as robustness against stragglers and single points of failure through decentralized, asynchronous, and concurrent operations. The third algorithm extends the second algorithm by enforcing robustness against uncertainties with the help of a time-tracking scheme. Through theoretical analyses, we establish the convergence properties of our algorithms and show that our decentralized algorithms achieve a computational complexity of $O(1)$ for each worker node, whereas our centralized algorithm achieves a computational complexity of $O(N)$ for the central node and $O(1)$ for each of the remaining nodes. Through numerical simulations with various settings, we show that our algorithms have converged significantly faster than several state-of-the-art ADMM algorithms with well-established convergence properties.

The final part of the thesis extends these optimizations to highly volatile systems characterized by message dropouts and dynamic topologies. Here, we introduce two algorithms to adapt ADMM to dynamic systems, where new nodes may be added amidst the process at the same time as the system may encounter issues with stragglers and message dropouts. The algorithms achieve fast convergence and flexibility by allowing nodes to choose a step size that is best suited for their own system and by allowing the nodes to move on to the next iteration even when not all nodes have made an update for the current iteration. More importantly, these algorithms incorporate a contribution tracking mechanism to ensure consistency despite message loss. Furthermore, these algorithms enforce robustness against uncertainties by removing the need to predefine the waiting time or the minimum number of updates before moving to the next iteration. Here, an approximation mechanism is also introduced to give stragglers more time to compute their variables while the faster nodes move on to the next iteration.

To summarize, this thesis provides accelerated algorithms for fast convergence in distributed optimization, with solutions tailored for both large-scale and dynamic networks.

Water is an essential resource, and providing reliable water supply to all citizens is a critical societal responsibility. In urban areas, this is typically achieved through water distribution pipe networks. Unfortunately, such networks are universally plagued by leakages, which account for substantial water losses, wasting both energy and financial resources while reducing supply capacity. Beyond these economic consequences, leaks pose serious hazards: they can undermine and damage infrastructure, and create risks of contamination and disease spread. The prompt localization and repair of leakages is therefore of paramount importance.

Hydraulic model-based leak localization is an approach that estimates leak positions by combining equations describing network hydraulics with sensor measurements of water production, consumption, pressure, and potentially flow. The field has a long history of academic research, with many sophisticated and effective methods having been proposed. This thesis aims to complement the existing literature, which largely relies on simulation-based performance evaluation, by establishing theoretical conditions under which leak localization can be guaranteed to succeed. The results are derived under idealized assumptions, yet they illuminate behavior that is observed in more general circumstances.

The thesis is based on five articles presented across four technical chapters, preceded by a modeling chapter that introduces the abstractions and assumptions adopted throughout. The modeling chapter also serves to unify notation and, notably, introduces a formulation for leakages that can occur anywhere along any pipe in the network, going beyond the common practice of restricting attention to node leakages only.

In the first technical chapter, a single-pipe scenario is considered. The leaking pipe has unknown hydraulic resistance parameters, and the problem of simultaneously estimating the uncertain model and localizing the leak is analyzed. Under a particular parameterization, the problem takes a bilinear form, for which dedicated solution methods are examined and convergence rates derived. Practical performance is further assessed through a simulation study.

In the second technical chapter, hydraulic resistance parameter estimation is considered; a problem that, much like leak localization, is often ill-posed. For a small-scale network representing a well-field for water production, sufficient conditions for unique parameter estimation are derived. These conditions highlight the necessity of adequate data variation, and a simulation example illustrates how estimation performs in practice.

In the third technical chapter, leak localization is investigated in a network with pipes connected in parallel. This configuration serves as a condensed representation that isolates the positional ambiguity introduced by loops. Necessary sensor conditions for leak localization are derived, and it is shown that even under maximal sensorization, unique determination of the leak position is often impossible or at least severely limited. The findings are validated through both simulation examples and physical laboratory experiments.

In the fourth and final technical chapter, leak localization in general networks is examined. The distinguishability and localizability problems---fundamental prerequisites for reliable leak localization---are studied, and structural sufficient conditions on network topology and pressure sensor placement are derived to guarantee both properties. Illustrative examples in small-scale networks are provided to elucidate the theoretical findings.

Taken together, the presented results offer rigorous insight into the requirements for reliable leak localization, while also pointing to open questions that remain to be addressed in future work.

Condensed matter physics studies how many quantum particles, when brought together, conspire to give rise to collective behaviours that are far richer than those of the individual particles themselves. This is the physics of emergence: the idea that a large system can have properties that are not obvious from its microscopic building blocks alone. The work presented in this thesis explores how to characterise the properties of such many-particle states from three interconnected points of view: topology, quantum transport, and quantum information. The common idea behind these themes is that the properties of quantum matter may be accessed directly from quantum states themselves, without necessarily referring to an underlying Hamiltonian, circuit, or protocol used to generate them.

Symmetry-protected topological phases of matter arise from the impossibility of continuously deforming certain quantum states into others whilst preserving their symmetries. In fermionic quantum matter, the phases resulting from such impossibility are those of topological insulators and superconductors. While having an insulating bulk, these materials possess metallic boundary states robust to perturbations preserving the protecting symmetries of their phase. Mathematically, the topological nature of these materials is characterised by the calculation of topological invariants, integer or half-integer numbers distinguishing between different topological phases. These invariants have been traditionally formulated for crystalline quantum matter, where one may use momentum space to perform mathematical calculations. This thesis focuses instead on characterising the topological phases of noncrystalline matter, which includes disorder and amorphicity. We develop a series of mathematical tools, known as local topological markers, which reformulate traditional topological invariants in real space and are therefore applicable even in the absence of crystallinity. Local topological markers are formulated in terms of the one-particle density matrix, a correlation function that only requires the knowledge of the quantum state in question to be calculated. Thus, the framework for topological characterisation we discuss, which also extends to certain higher-order topological phases, is purely state-based: it does not need to specify any parent Hamiltonian nor protocol used to prepare the quantum state to be characterised.

Contrary to the classical picture of electricity where electrons behave as solid spheres drifting through a wire, the electronic transport through a nanostructure is shaped by quantum effects like tunnelling and interference. Quantum transport studies how these effects determine the conducting properties of mesoscopic samples---those that are neither microscopic nor fully macroscopic, but lie in an intermediate regime. In this thesis, we use quantum transport to study topological insulator nanowires, where the interplay between topology, deviations from crystallinity, and symmetry constraints shape electronic transmission. The characteristic feature of topological nanowires is the appearance of a perfectly transmitted mode when a certain magnetic flux is threaded through the wire's cross section. Varying the magnetic flux interpolates from perfect transmission to situations where electronic interference is enhanced, suppressing transmission. This is seen in the behaviour of conductance, which oscillates as a function of magnetic flux in what is known as Aharanov-Bohm oscillations. The mechanism originating the perfectly transmitted mode in a topological nanowire relies on time-reversal symmetry appearing at specific values of the magnetic flux. In amorphous nanowires,  where time-reversal symmetry may be broken at any magnetic flux, the perfectly transmitted mode loses its protection, and the characteristic conductance oscillations of a topological nanowire may be lost. We study how the perfectly transmitted mode may be protected even when time-reversal symmetry is broken, and relate its loss to a topological phase transition driven by amorphicity.

Entanglement is a distinctive feature of condensed matter systems: the possibility that quantum particles may be correlated in a way that has no classical analogue. Although inherently quantum-mechanical, entanglement alone is not a sufficient ingredient for quantum computation to outperform classical computers. The reason is that there exists a special class of quantum states, stabiliser states, that may be highly entangled whilst admitting an efficient classical representation---calculations involving stabiliser states are tractable in modern-day computers. The deviation of a quantum state from the set of stabiliser states goes by the name of quantum nonstabiliserness or magic, and, in the same way as entanglement, it is not a guarantee of quantum complexity on its own. In this thesis, we develop a method to characterise quantum states that combines entanglement and nonstabiliserness. To do so, we use the framework of the information lattice, whichsystematically decomposes the correlations in a quantum state in terms of their scale---the spatial extent of correlations. Through the information lattice, we distinguish different contributions to the nonstabiliserness content of a quantum state, which may have either a single-particle origin or come from intrinsic many-particle correlations. This framework, which is purely state-based, provides a first step to a fully scale-resolved description of nonstabiliserness and contributes to the characterisation of quantum states in terms of their complexity.

Sample collection and processing are fundamental components in the study of a multitude of biological systems. The reliability of downstream analyses depend greatly on the integrity of the collected material and the processing steps. Consequently, efforts have been directed to the refinement of sampling strategies to collect samples while still maintaining quality, reducing invasiveness to allow for longitudinal and frequent monitoring, and enable decentralized sampling, simplifying the overall workflow. Thus, these are approaches that reduce the logistical burden and enable more decentralized, or field-based collection approaches that can broaden the context of where biological data can be obtained. 

This thesis presents novel microfluidic devices, designed to simplify sample collection and processing, for both plant and human systems, with the aim to detect environmental stressors and biomarkers of disease. Through the reduction of procedural complexity this work aim to support more accessible and easier sampling approaches that could be applied in different biological and health care settings. 

This is done through the development of a plant sap sampling device that is designed to enable low invasive sampling in-field. The device presents a disposable system with integrated drying, inspired by the dried blood spot concept for human diagnostics to enable collection and storage of a sap sample without the need of a cold chain. This device is also further developed and combined with a biosensor for detection of abscisic acid, a phytohormone used for stress signaling in plants, thus presenting on-site detection capabilities of drought and environmental stress. Additionally, the device is used for longitudinal monitoring of plants in an ozone environment where machine learning was used to identify predictors of ozone exposure in the phytohormonal fingerprint. 

The work also includes a device for a different biosample, blood. Here an autonomous capillary microfluidic filtration device for whole blood sample processing to extract nanoparticles and extracellular vesicles is presented. The device is based on a dual filtration approach to directly process a whole blood sample into a nanofiltrate. The device is characterized with polystyrene beads, synthetic extracellular vesicles and liposomes to verify functionality and is further used to evaluate cancer patient samples to detect biomarkers of disease. 

Together, these contributions present technological progress in the field of on-site sampling for plant sciences and a novel means of blood filtration for the analysis and detection of disease biomarkers. Thus, these contributions provide low invasive, patient and plant centric sampling approaches for potential applications in environmental monitoring, nutrient analysis, disease and treatment tracking or for screening purposes. 

This thesis investigates the radiation response of 4H-SiC bipolar junction transistors and integrated TTL inverter circuits under gamma irradiation, with emphasis on the role of bias conditions, dose rate, temperature, and oxide/interface processing. The work combines in-situ irradiation experiments using a ⁶⁰Co source, device- and circuit-level electrical characterization, and compact modelling approaches to establish a consistent understanding of radiation-induced degradation mechanisms in this technology.

The results show that the dominant degradation mechanism in the studied devices is an increase in base current, leading to a reduction in current gain. This behaviour is attributed to radiation−induced charge trapping in oxide layers and the formation of interface states at the SiC/SiO₂ boundary, which enhance surface recombination. Despite the wide bandgap of 4H-SiC and its intrinsically low carrier concentration, the radiation response is governed primarily by interface−controlled processes rather than by bulk material properties.

A key finding is the dependence of degradation on the electrical bias applied during irradiation. Passive or zero-bias conditions are shown to produce more severe degradation than active bias configurations. 

Circuit-level experiments demonstrate that transistor degradation propagates into inverter behavior through shifts in transfer characteristics and changes in supply current. While circuits may remain operational after irradiation, their functional margins are reduced and become strongly dependent on the bias conditions. 

The thesis also introduces a compact modelling framework that links radiation-induced physical mechanisms to SPICE-compatible parameters, enabling predictive circuit-level simulations under irradiation conditions.

The overall conclusion is that radiation hardness in 4H-SiC bipolar electronics in many cases can be much higher compared to silicon devices, but the radiation hardness should be understood as a condition-dependent system property, rather than as an intrinsic material characteristic.

When a person says "put that over there" while pointing at a shelf, the meaning depends on the spatial relationship between speaker, listener, and shared physical scene. Embodied agents that participate in such interactions must both produce spatially grounded gestures and interpret multimodal references. Yet these capabilities have largely been studied in isolation, with separate data, methods, and evaluation paradigms.

This thesis argues that gesture generation and referential grounding are two sides of the same communicative process, and that studying them jointly reveals structure that neither subfield surfaces alone. The argument is developed across seven papers. On the production side, contrastive speech-motion pretraining enables semantically aware co-speech gesture generation, while reinforcement learning with adversarial motion priors produces pointing gestures that are both spatially accurate and motorically natural, outperforming supervised baselines in a human referential identification study. A flow-matching architecture further combines semantic and spatial conditioning within a single generative system through distinct pathways.

On the comprehension side, the thesis introduces multimodal conversational datasets recorded in virtual reality and with wearable AR sensors, combining full-body motion, gaze, speech, and 3D scene context. Experiments show that state-of-the-art vision–language models fail on conversational references not for lack of perceptual capability, but because they cannot determine what is being referred to from underspecified language. A rewrite-based decoupling experiment isolates this bottleneck: once the referent is explicitly described, even simple detectors localize it accurately.

A central finding across both threads is that semantic reasoning, what is being communicated, and spatial reasoning, where it is directed, benefit from separate architectural treatment. On the production side, audio conditioning drives gesture timing while spatial targets determine direction; on the comprehension side, linguistic reasoning identifies the referent while visual perception localizes it. In both cases, architectures that maintain this separation outperform those that conflate heterogeneous signals into a shared representation. A shared data infrastructure, built incrementally across the papers, makes this parallel empirically testable: the same referential annotations that define conditioning targets for generation also define evaluation targets for grounding.

The thesis contributes methods, datasets, benchmarks, and evaluation protocols that support a unified view of spatially grounded communication in embodied agents, where producing and interpreting meaning are coordinated processes grounded in language, body, and shared physical space.

Biotechnology is a knowledge-intensive industry characterized by long

development timelines, high uncertainty, and substantial resource

requirements. Firms in this sector often depend heavily on external financial,

institutional, and knowledge-based resources to sustain research and

development and progress toward commercialization. Despite strong public

support, biotechnology firms in Sweden continue to face persistent resource

constraints that shape their growth trajectories and strategic choices. This thesis

examines how biotechnology firm access, mobilize, and manage critical

resources for growth within Sweden’s clustered and institutionally structured

innovation environment. The thesis comprises four complementary studies that

provide a multi-level analysis of biotechnology firm development. The first study

quantitatively examines how access to external financing and locating in cluster

areas influence firm survival. The second study explores how biotechnology

incubators strategically build and manage networks to secure essential

resources for early-stage ventures. The third study investigates how

biotechnology entrepreneurs perceive and navigate evolving resource

constraints and institutional support across different development stages. The

fourth study analyzes how intellectual capital and innovation activities shape

biotechnology firms’ capital structure decisions, particularly their reliance on

equity versus debt financing.

The findings show that biotechnology firm development is shaped not only by

resource availability but also by how resources are coordinated and strategically

mobilized over time. Public funding supports early value creation, incubatorsfunction as active resource coordinators, entrepreneurs adapt strategies as

resource needs evolve, and financing decisions reflect pragmatic trade-offs

related to innovation and intangible assets. Overall, the thesis contributes to

research on biotechnology entrepreneurship by integrating financial,

institutional, and strategic perspectives and offers insights for policymakers,

incubator managers, and entrepreneurs in science-based industries.

Throughout life, cells undergo cell-state transitions, such as changing cell identities or adapting to environmental stress. Because cells in an organism contain the same set of genes, distinct cell-state transitions arise from differences in gene expression. Gene expression is primarily regulated at the level of transcription, a complex process driven by RNA Polymerase II (Pol II). Beyond Pol II, transcription is coordinated by numerous transcription factors and at regulatory genomic regions, such as promoters and enhancers. By tracking transcription and its regulation upon cell-state transitions, basic biological principles can be identified and leveraged when studying diseases.

The papers presented in this thesis trigger and track transcriptional responses upon two types of cell-state transitions: stress and differentiation. Differentiation is a slow, controlled process that permanently alters the transcriptional program. In contrast, stress induces rapid, transient changes to the transcriptional program, keeping the cell alive and enabling recovery once the stress ends. A classic example of a stress response is the heat shock response (HSR), which is activated by elevated temperatures and other stressors that cause protein misfolding. The HSR represses thousands and activates hundreds of genes, some of which encode chaperones that assist in the correct folding of other proteins.

Across the papers, cell-state transitions are triggered mainly by elevated temperatures or chemical components. After a trigger, the transitions are explored primarily with mRNA-seq and PRO-seq, two genome-wide sequencing techniques. mRNA-seq tracks steady-state levels of mature mRNA, while PRO-seq tracks transcription of nascent RNAs. Paper I provides a computational workflow to generate maps of functional genomic regions, such as promoters and enhancers, from mapped PRO-seq reads. Such maps are central to studying transcriptional responses, and the workflow was used to generate results in Papers II–IV. In Paper II, transcriptional responses and master regulators that drive differentiation in human erythroid cells are explored. A brief signal causes slowly propagating and long-lasting transcriptional changes that prepare cells for future tasks involving oxygen transport. The paper also provides the first direct comparison between induced differentiation and stress in a cell model, revealing drastically distinct coordination of transcriptional responses.

Paper III–V focus on transcriptional responses upon stress. In Paper III, transcription in golden retriever macrophages is tracked under normal conditions and upon heat stress. The first complete transcriptional profiling of the dog genome and characterization of the HSR in dog are provided. In Paper IV, transcription is tracked in fly, mouse, dog, and human upon heat stress. The paper uncovers that intron length is an evolutionarily conserved regulatory mechanism that times the production of heat-induced chaperones. This finding addresses the broader question of why any genome has introns that span hundreds of kilonucleotides. In Paper V, transcriptional programs of three distinct forms of protein-damaging stressors are compared: heat shock (HS), HSP90 inhibition, and polyglutamine (polyQ) aggregation. PolyQ aggregation is a feature of Huntington’s disease (HD), for which no cure exists. The paper reveals that transcriptional responses are fundamentally different for the three stressors. PolyQ aggregation is linked to impaired acute stress responses, which could explain why previous attempts to ameliorate the disease by inducing the HSR have not been successful. There is also a lack of commonly induced or repressed genes across mouse brain tissues under polyQ stress, which suggests that therapeutics targeting pathways rather than specific genes should be explored. Overall, this thesis expands current knowledge of transcriptional responses upon cell-state transitions in health and disease.

Within the industry, quantitative safety analysis is often based on well-established methods that have existed for decades. The perhaps most prominent example is fault trees, in which the probability of a system failure is computed from the probability of component-level malfunctions. While these classical methods has the advantage of being well-established and easy to understand, they are lacking in two major areas. Firstly, the models does not describe the architecture of the system. Since this is the case, they are error-prone when changes are made in the system and two different engineers tend to produce vastly different models of the system. Secondly, they only support exponential distributions as a mean to introduce stochastic behavior in the models. As a result of this restriction, the complex dynamic behavior of the cyber-physical system that constitutes a road vehicle today cannot be modeled accurately. Within the academia, many methods, languages, and tools have been suggested in the last decades that would would help circumvent one or both of these restrictions. However, these methods has to date not reached prominent traction within the industry. In this thesis, languages and analysis methods with tool support for quantitative safety analysis that surpass the above mentioned restrictions while still being attractive candidates for the industry are presented.

In traditional data management systems, queries have well-defined semantics and produce exact results. Integrating Machine Learning inference into data processing pipelines disrupts both properties by introducing operators whose outputs are approximate rather than exact. This thesis establishes two foundations for trustworthy AI-native data systems: empirical characterization of ML operator execution cost, and formal, declarative correctness guarantees that the system enforces on behalf of the user. We develop these foundations across three levels of abstraction, from single-operator cost, to single-operator correctness, to their joint optimization at the pipeline level, and establish Conformal Prediction as a practical statistical foundation for this approach. We introduce Crayfish, a benchmarking framework for ML inference within dataflow engines that reveals how interactions between serving tools, stream processors, and pipeline configurations shape inference costs in ways that are difficult to anticipate from component-level behavior alone. We propose ConANN, the first framework to provide distribution-free recall guarantees for Inverted File-based Approximate Nearest Neighbor search, using conformal methods to replace heuristic index tuning with formal statistical guarantees. At the pipeline level, we study joint cost and correctness optimization in the context of Neural Graph Databases, where multi-hop queries over Knowledge Graphs interleave retrieval and neural execution. We formalize a hybrid query optimization architecture for this setting, then introduce ConRAD, which enforces end-to-end recall guarantees for multi-hop queries while dynamically bypassing expensive neural inference when recall targets can be met with local graph evidence alone. Taken together, these contributions show that the rigor users expect from traditional data systems need not be abandoned as those systems become increasingly driven by Machine Learning.

The rapid growth of data-intensive applications such as edge computing, artificial intelligence and the Internet of Things is pushing the limits of conventional CMOS electronics. In these systems, static leakage currents increasingly dominate power consumption. Nanoelectro-mechanical (NEM) switches are promising candidates for beyond-CMOS electronics due to their near-zero off-state leakage, abrupt switching characteristics, and robustness under extreme operating conditions, offering a route to dramatically reduce static power dissipation in future integrated circuits. However, practical NEM-based systems require scalable device architectures, reliable switch contacts, and CMOS-compatible integration strategies. This thesis addresses these challenges through the realization and integration of a CMOS-compatible NEM switch device library within commercial CMOS foundry platforms. The work investigates three complementary NEM switch architectures for logic and memory applications: a volatile three-terminal (3-T) switch, a volatile four-terminal (4-T) switch with decoupled actuation and signal paths, and a non-volatile seventerminal (7-T) switch. Building upon concepts established in earlier research within our group, the 3-T and 7-T devices are miniaturized and optimized through systematic studies of beam geometry and contact materials for low-voltage operation and improved switching behavior. A major contribution of this thesis is the optimization and experimental realization of the 4-T architecture, enabling body-bias-assisted reduction of the pull-in voltage and advanced circuit configurations. Two CMOS-compatible integration approaches are developed and experimentally validated: (1) Monolithic integration within the IMEC iSiPP50G silicon photonics SOI foundry platform, and (2) heterogeneous 3-D integration within the X-FAB XI10 SOI CMOS process. The first method enabled co-fabrication of all three NEM switch architectures on a single commercial foundry chip for the first time. Electrical characterization confirms volatile switching in the 3-T and 4-T devices, pull-in voltage reduction in the 4-T switch through body biasing, and both volatile and nonvolatile operation in the 7-T switch through contact engineering. However, in this approach, circuit scalability is limited by routing density inherent to planar integration, while Au contact stiction constrains switch reliability. The second approach addresses these limitations by vertically integrating the NEM device layer above the completed back-end-of-line (BEOL) through heterogeneously 3-D integration. This architecture alleviates routing constraints and improves device reliability using Ruthenium (Ru) switch contacts. Ru-coated devices demonstrate substantially improved cycling endurance, and a complementary inverter implemented with Ru-coated 3-T switches validates the feasibility of functional BEOL-integrated NEM circuits.

Somatic freedoms are ethical commitments to bodies as whole, living and corporeal subjectivities within processes of research and design. They view ethical sensibilities as grounded in our bodies and felt experiences, and intimately shaped by our interpersonal, socio-cultural, and political relations. This position argues that embodied ethical knowledge is implicit in many aspects of research and design practice, as well as our interactions with machines. I explore felt experiences of ethics across three multidisciplinary art and design collaborations: (i) Robots, Dance, Different Bodies, (ii) Embrace Angels, and (iii) How to Train Your Drone. Across these projects, I investigate how misalignments between ethical sensibilities can invite deeper modes of allyship; how frictions between sensibilities can be engaged to forge bridges between interdisciplinary practitioners; how sensibilities are practiced within human-machine interactions; and how we might design machines that can support bodies in uncovering potential pathways to ethical transformation. 

Across these projects, I chart the development of somatic freedoms as ethical commitments to bodies and the vibrant, material and corporeal pursuit of change and transformation. These are commitments to creating conditions in which transformation is supported, enhancing the possibilities available to bodies to achieve change, facilitating the processes through which bodies transform their relations with the world, and by extension, change how they are constituted within those relations. My papers demonstrate processes of human-machine ethical reconfiguration: the ongoing reorganisation of relations between bodies and machines through which ethical transformation can be realised. These processes require traversing the boundaries of our bodies within our relational engagements with the world around us, experiencing ourselves as subjects who move but also objects that are moved within those encounters. This is a process by which bodies can engage in the work of ethics, seek out better ways of relating to each other, meet new ethical challenges, and move towards conditions of flourishing.