http://hdl.handle.net/10900/42133 Sun, 02 Aug 2026 06:46:48 GMT 2026-08-02T06:46:48Z http://hdl.handle.net/10900/182019 Integrative Computational Analyses Across the Central Dogma: Developing Applications for Prokaryotic Characterization in the Context of Microbial Control Witte Paz, Mathias Alexander The global rise in antibiotic resistance, coupled with the declining discovery of new compounds, has created a crisis that demands innovative strategies for microbial control. Identifying such novel strategies requires a mechanistic understanding of bacterial molecular pathways. The central dogma of molecular biology, expanded by newer findings on the flow of biological information, provides a fundamental roadmap for prokaryotic characterization based on high-throughput experiments and computational approaches. However, capturing this complexity requires integrative and reproducible computational frameworks that enable robust cross-layer analyzes. This thesis addressed this challenge by developing and applying computational methods that support integrative and visual analyses across the central dogma, with focus on microbial characterization for their control. To enhance the interactive characterization of prokaryotic genomes, the visual analytics tool Evidente is introduced, with the goal of bridging the gap between the exploration of single nucleotide polymorphisms (SNP) and the evolutionary context. Unlike traditional visualization tools, Evidente classifies SNPs based on clade-specificity and enables their visualization with metadata, as well as linking them to a functional context. By applying it to bacterial pathogens, this approach demonstrated how genome-scale variation can be interpreted in an evolutionary context to generate functional and phenotypic hypotheses. Recognizing that genomic data alone are insufficient to explain phenotypic diversity, this thesis integrated transcriptomic data through two complementary web-applications: TSSpredator-Web and TSS-Captur. They address the characterization of the transcriptome's architecture and link the genomic with the transcriptomic layer of the central dogma. TSSpredator-Web extends the tool TSSpredator to identify and classify transcription start sites (TSS), and allows exploration of genome-wide TSS maps together with genomic data. Based on TSS maps, the Nextflow-based pipeline TSS-Captur characterizes transcripts starting from unclassified TSS via computational methods for sequence classification, termination site prediction, and analyses of secondary structure and promoter regions. Together, both approaches showed how transcriptomic data can be used for annotation refinement and transcript discovery, bridging the genomic and transcriptomic layer. Lastly, the power of integrative analyses is demonstrated through two systematic studies of bacterial metallophore systems. The first project focused on the characterization of known metallophore mechanisms across the Staphylococcal genus using computational analyses based on genomic data, representing the first comprehensive genus-wide analysis of these systems. Expanding on this topic, the second study investigates how nasal Corynebacterium species exploit metallophores synthesized by Staphylococcus aureus, suggesting a novel way of controlling pathogens. This was achieved by identifying structural homologs of lipoproteins through the development of the Nextflow-based pipeline PRESERVE, and contextualizing the findings with transcriptomic data. By analyzing the putative promoter regions, we gained insight into how these mechanisms are regulated in Corynebacteria. These studies illustrate how the integration of multiple layers of biological data can shed light into the interactions between species, providing knowledge that can be translated into interference of colonization strategies, and be exploited for microbial control. In summary, this thesis describes a methodological framework for integrative computational analyses across the central dogma. By combining reproducible approaches with interactive visual exploration, it illustrates how integrating data from different biological layers can help to generate insights for prokaryotic characterizations. These insights can be used to establish innovative interventions for microbial control, offering a computational pathway to address the escalating challenges of antibiotic resistance.; Der weltweite Anstieg der Antibiotikaresistenz in Verbindung mit der sinkenden Entdeckungsrate neuer Wirkstoffe hat zu einer internationalen Krise geführt, die innovative Strategien zur Bekämpfung von bakteriellen Pathogenen erfordert. Um neue Strategien zu identifizieren, ist es notwendig, die komplexen Mechanismen hinter den molekularen Signalwegen von Prokaryoten zu verstehen. Das zentrale Dogma der Molekularbiologie, erweitert durch neuere Erkenntnisse über den Fluss biologischer Informationen, liefert einen grundlegenden Fahrplan für die prokaryotische Charakterisierung auf der Grundlage von Hochdurchsatz-Experimenten und computergestützten Ansätzen. Um diese Komplexität zu erfassen, sind jedoch integrative Methoden erforderlich, die explorative und reproduzierbare Analysen ermöglichen, um über mehrere Ebenen Erkenntnisse zu gewinnen. Diese Dissertation hat sich dieser Herausforderung gestellt, indem sie integrative und visuelle Analysen über das zentrale Dogma hinweg entwickelt und durchführt. Dafür wurden computergestützte Methoden entwickelt und angewendet, wobei der Schwerpunkt auf der mikrobiellen Charakterisierung für deren Kontrolle liegt. Um die interaktive Charakterisierung prokaryotischer Genome zu verbessern, wurde das Visual Analtytics Tool Evidente eingeführt, mit dem Ziel, die Lücke zwischen der Erforschung von Einzelnukleotid-Polymorphismen (SNP, engl. single nucleotide polymorphism) und dem evolutionären Kontext zu schließen. Im Gegensatz zu herkömmlichen Visualisierungstools klassifiziert Evidente SNPs auf der Grundlage der Kladenspezifität und ermöglicht ihre Visualisierung mit Metadaten sowie ihre Verknüpfung mit funktionalen Annotationen. Durch die Anwendung auf bakterielle Pathogene zeigte dieser Ansatz, wie genomweite Variationen in einem evolutionären Kontext interpretiert werden können, um robuste funktionelle und phänotypische Hypothesen zu generieren. Ausgehend von der Erkenntnis, dass Genomdaten allein nicht ausreichen, um die phänotypische Vielfalt zu erklären, untersuchte diese Arbeit die Integration von Transkriptomdaten durch zwei sich ergänzende Webanwendungen: TSSpredatorWeb und TSS-Captur. Diese Tools befassen sich mit der Charakterisierung der Transkriptomarchitektur und verbinden die Genom- mit der Transkriptomebene des zentralen Dogmas. TSSpredatorWeb erweitert das Tool TSSpredator zur Identifizierung und Klassifizierung von Transkriptionsstartstellen (TSS) und ermöglicht die Untersuchung genomweiter TSS-Karten zusammen mit Genomdaten. Basierend auf TSS-Karten, die auf Nextflow-basierte Pipeline TSS-Captur charakterisiert Transkripte, die von nicht klassifizierten TSS stammen, indem es Methoden für die Sequenzklassifizierung, die Vorhersage von Transkriptionsterminatoren, die Sekundärstrukturanalyse und die Promotoranalyse integriert. Zusammen zeigten beide Ansätze, wie Transkriptomdaten für die Verfeinerung von Annotationen und die Entdeckung von Transkripten verwendet werden können. Dadurch wird eine Brücke zwischen der genomischen und der transkriptomischen Ebene geschlagen. Schließlich wird die Leistungsfähigkeit integrativer Analysen anhand von zwei systematischen Studien zu bakteriellen Metallophorsystemen demonstriert. Das erste Projekt konzentrierte sich auf die Charakterisierung bekannter Metallophormechanismen in der Gattung Staphylococcus mithilfe computergestützter Analysen auf der Grundlage genomischer Daten und stellte die erste umfassende gattungsweite Analyse dieser Systeme dar. In Erweiterung dieses Themas untersuchte die zweite Studie, wie nasale Corynebacterium-Arten die von Staphylococcus aureus synthetisierten Metallophore nutzen, und deutet damit einen neuen Weg zur Bekämpfung von Krankheitserregern an. Hierzu wurden strukturelle Homologe von Lipoproteinen durch die Entwicklung der Nextflow-basierten Pipeline PRESERVE identifiziert und die Ergebnisse mit Transkriptomdaten in Zusammenhang gebracht. Durch die Analyse der mutmaßlichen Promotorregionen wurden Einblicke in die Regulation dieser Mechanismen in Corynebacteria gewonnen. Diese Studien zeigen, dass die Integration mehrerer Ebenen biologischer Daten Aufschluss über die Wechselwirkungen zwischen Arten geben kann. Diese Erkenntnisse lassen sich auf Kolonisationsstrategien übertragen und können somit für die mikrobielle Kontrolle genutzt werden. Zusammenfassend beschreibt diese Dissertation einen methodischen Rahmen für integrative computergestützte Analysen über alle Ebenen des zentralen Dogmas hinweg. Durch die Kombination reproduzierbarer Ansätze mit interaktiver visueller Exploration wird veranschaulicht, wie die Integration von Daten aus verschiedenen biologischen Ebenen dazu beitragen kann, Erkenntnisse für die Charakterisierung von Prokaryoten zu gewinnen. Dieser computergestützte Weg ermöglicht es, Erkenntnisse zu gewinnen, mit denen innovative Maßnahmen zur mikrobiellen Kontrolle entwickelt werden können, um den zunehmenden Herausforderungen der Antibiotikaresistenz zu begegnen. Fri, 31 Jul 2026 00:00:00 GMT http://hdl.handle.net/10900/182019 2026-07-31T00:00:00Z http://hdl.handle.net/10900/182018 Privacy by Design: From Distributed Learning to Post-Deployment Risks Swaminathan, Arjhun Modern data processing activities increasingly rely on sensitive data, from medical images and genomic data to electronic health records, to enable research, discovery, diagnosis, and decision-making. Yet the very capabilities that make these systems powerful also create privacy risks: data must often be shared across institutions for generalization and accuracy, and once models and datasets are released for downstream use, they become long-lived artifacts that others can query, link, or exploit. Privacy by design, the principle that privacy should be a foundational property of any data processing activity rather than a post-incident patch, offers a response to these risks. However, translating this into concrete technical practice remains a challenge, because the right answer depends on where in the processing lifecycle one stands, what is being protected, and what assumptions about adversaries are relevant. This thesis approaches privacy by design as a technical agenda organized around two regimes: pre-deployment and post-deployment. On the pre-deployment side, we develop privacy-preserving methods for computation on distributed data under semi-honest threat models. We introduce randomized-encoding based approaches for scalable kernel learning on medical images and for multi-site genome-wide association studies on quantitative phenotypes. We further show that widely used classical kernels can be realized through quantum feature maps, and introduce a distributed secure quantum architecture for kernel computation, validated on simulated quantum hardware. On the post-deployment side, we study what deployed artifacts reveal and how that exposure can be exploited or mitigated. We introduce a targeted adversarial attack for hard-label black-box image classifiers that leverages edge information from images to accelerate attack progress under strict query budgets, consistently outperforming existing methods in the low-query regime across diverse architectures. We also develop a topological framework for adaptive k-anonymisation for dynamic datasets, enabling incremental updates to anonymised data releases without full recomputation when the underlying data changes. Taken together, the results presented in this thesis demonstrate that privacy by design for data processing activities is not a single technique but a discipline whose realization spans a composition of architectures, threat models, and lifecycle stages. Fri, 31 Jul 2026 00:00:00 GMT http://hdl.handle.net/10900/182018 2026-07-31T00:00:00Z http://hdl.handle.net/10900/182002 Technological Advances in Bioreporter-Based Screening for Mechanism-Informed Antibiotic Discovery in Bacillus subtilis Schubert, Julian Frederik The discovery of new antibacterial agents with potent and well-defined mechanisms of action has become essential for addressing the growing burden of antibiotic resistance. However, traditional screening pipelines remain constrained by low throughput, limited specificity, and labor-intensive dereplication. These limitations highlight the need for new screening strategies that accelerate antibacterial discovery and enable early mechanistic assessment. This dissertation describes the advancement of a bioreporter-based screening platform in Bacillus subtilis that combines mechanism-informed whole-cell screening with an efficient compound discovery and dereplication workflow. For this purpose, a new generation of bioreporters based on the bacterial luciferase system was developed, including a novel bioreporter specific for proteotoxic stress. In parallel, a customized workflow for the discovery and rapid dereplication of antibacterial agents was established, enabling the seamless integration of the bioreporter technology. The first study implemented the compound-resolved bioactivity-based metabolomics pipeline, which combines the bioreporter panel with high-frequency microfractionation onto microfluidic paper-analytical devices and non-targeted LC-MS/MS. This approach enabled high-throughput antibiotic screening and the identification of bioactive compounds from pure compounds, crude extracts, and producer strains, while providing early insights into their mechanisms of action. The second study expanded the mechanistic coverage of the bioreporter panel by developing a sensitive bioreporter that signals proteotoxic stress caused by the accumulation of damaged or misfolded proteins. Validation with an extensive set of antibacterial reference compounds confirmed the high specificity of the bioreporter for antibacterial agents that induce proteotoxic stress and enabled the discovery of several compounds not previously associated with this mechanism. Integration of the bioreporter with the microfractionation workflow facilitated the identification of an extensive molecular network of streptothricin derivatives from the Tübingen collection of actinomycete producer strains, including multiple putatively novel analogues. Overall, this work establishes a versatile, bioreporter-based screening platform for mechanism-informed antibiotic discovery and dereplication. By directly linking bioactivity to compound identity at early stages of screening, this approach enables efficient exploration of natural products and the discovery of new antibacterial agents. Fri, 31 Jul 2026 00:00:00 GMT http://hdl.handle.net/10900/182002 2026-07-31T00:00:00Z http://hdl.handle.net/10900/181936 A Study of 3D Representations for Human–Scene Interaction: From Classical to Neural Mir, Mohamad Aymen Creating virtual humans that look realistic and behave naturally in 3D environments is a central challenge in computer vision and graphics. At its core lies a fundamental question: how should we represent 3D humans and their environments to enable realistic interaction? Classical mesh and point cloud representations provide the geometric machinery for physics, collision detection, and path planning—but achieving photorealism requires complex material models, global illumination, and high polygon counts. 3D Gaussian Splatting offers photorealistic rendering at real-time rates from images alone—but provides no explicit surfaces, distance fields, or navigable structures. Can these missing geometric capabilities be recovered entirely from Gaussian fields? This thesis investigates this question in two parts: the first leverages the geometric strengths of classical representations for human appearance, capture, and motion synthesis, while the second demonstrates that the same capabilities can be extracted directly from raw Gaussian fields, enabling photorealistic human–scene interaction without classical geometry. The broader question of how to represent 3D humans is especially pressing as we enter the era of video diffusion models, which can generate photorealistic animations of humans from 2D data alone, challenging the very relevance of 3D representations. While the challenge is valid, we contend that 3D representations will remain essential for simulators, game engines, and digital twins, where geometry-consistent rendering, multi-view coherence, and explicit control over motion, lighting, and physics are required—capabilities that 2D generative models fundamentally cannot provide. But we also contend that for 3D representations to remain competitive, they must close the photorealism gap: the answer is not to abandon 3D, but to evolve it. Neural 3D representations can deliver both geometric control and photorealism. This thesis is our contribution toward that goal. In the first part, we use classical representations to address foundational problems. We introduce Pix2Surf, a method to transfer textures of clothing images to 3D garments worn on top of SMPL in real time, learning dense correspondences from garment silhouettes to UV maps using shape information alone. We then introduce HPS (Human POSEitioning System), a method to recover the full 3D pose of a human registered with a 3D scan of the surrounding environment using body-mounted sensors. HPS fuses camera-based self-localization with IMU-based body tracking, enabling capture across environments spanning 300–2500 m². Next, we introduce a method for synthesizing animator-guided human motion across 3D scenes by composing short-term motions in a canonical coordinate frame, generating long sequences of diverse actions without scene-specific training data. In the second part, we introduce 3D Gaussian Splatting as a unified representation for photorealistic human–scene interaction. We present InteractSplat, an end-to-end pipeline that reconstructs separate controllable human and object models from multi-view video and animates them together in diverse 3DGS environments, enabling long-horizon sequences where avatars navigate scenes, pick up objects, and set them down at new locations. We then present SALA, the first method to synthesize human interactions in diverse 3D environments using 3DGS as the sole underlying representation, extracting navigable structures from raw Gaussian fields and introducing differentiable contact refinement in Gaussian space. To improve the realism of composited avatars, we introduce RAGA, a ray-traced shadow casting formulation that computes physically plausible shadows entirely in Gaussian space at interactive rates. Finally, we present AHOY, which exploits the fact that 3DGS can be reconstructed from images alone—without the dense multi-view capture or depth sensors that mesh-based avatars require—to build complete, animatable avatars from in-the-wild YouTube video despite heavy occlusion. Using video diffusion priors, AHOY unlocks YouTube-scale footage as a practical source of 3D human assets that can be animated and composited into 3DGS scenes. Together, the contributions of this thesis demonstrate that 3DGS can serve as a complete replacement for classical representations in the human–scene animation pipeline—from navigation and locomotion to contact modeling and shadow casting—while delivering superior visual fidelity. Wed, 29 Jul 2026 00:00:00 GMT http://hdl.handle.net/10900/181936 2026-07-29T00:00:00Z