Colorectal cancer is one of the leading causes of cancer related deaths in western world. While most common site for metastasis for colon cancer is liver, lung, and the peritoneum, metastasis to various other organs such as brain, bones and thyroid has been reported. Metastatic lesions to the small bowel are more common than primary lesions and most common primary neoplasms that metastasize to the duodenum are lung cancer, renal cell carcinoma, breast cancer, and malignant melanoma. We report a very rare case of recurrent adenocarcinoma of colon metastasizing to duodenum after 2 years of curative resection of primary cancer. Surgical resection for curative intent as well as palliative management is recommended.

Information mining and knowledge learning from sequential data is a field of growing importance in both industrial and academic fields. Sequential data, which is the natural representation format of the information flow in many applications, usually carries enormous information and is able to help researchers gain insights for various tasks such as airport threat detection, cyber-attack detection, recommender system, point-of-interest (POI) prediction, and citation forecasting. This dissertation focuses on developing the methods for sequential data-driven applications and evolutionary dynamics characterization for various topics such as transit service disruption detection, early event detection on social media, technology opportunity discovery, and traffic incident impact analysis. In particular, four specific applications are studied with four proposed novel methods, including a spatiotemporal feature learning framework for transit service disruption detection, a multi-task learning framework for cybersecurity event detection, citation dynamics modeling via multi-context attentional recurrent neural networks, and traffic incident impact forecasting via hierarchical spatiotemporal graph neural networks. For the first of these methods, the existing transit service disruption detection methods usually suffer from two significant shortcomings: 1) failing to modulate the sparsity of the social media feature domain, i.e., only a few important ``particles'' are indeed related to service disruption among the massive volume of data generated every day and 2) ignoring the real-world geographical connections of transit networks as well as the semantic consistency existing in the problem space. This work makes three contributions: 1) developing a spatiotemporal learning framework for metro disruption detection using open-source data, 2) modeling semantic similarity and spatial connectivity among metro lines in feature space, and 3) developing an optimization algorithm for solving the multi-convex and non-smooth objective function efficiently. For the second of these methods, the conventional studies in cybersecurity detection suffer from the following shortcomings: 1) unable to capture weak signals generated by the cyber-attacks on small organizations or individual accounts, 2) lack of generalization of distinct types of security incidents, and 3) failing to consider the relatedness across different types of cyber-attacks in the feature domain. Three contributions are made in this work: 1) formulating the problem of social media-based cyber-attack detection into the multi-task learning framework, 2) modeling multi-type task relatedness in feature space, and 3) developing an efficient algorithm to solve the non-smooth model with inequality constraints. For the third of these methods, conventional citation forecasting methods are using the traditional temporal point process, which suffers from several drawbacks: 1) unable to predict the technological categories of citing documents and thus are incapable of technological diversity assessment, and 2) require prior domain knowledge and thus are hard to extend to different research areas. Two contributions are made in this work: 1) formulating a novel framework to provide long-term citation predictions in an end-to-end fashion by integrating the process of learning intensity function representations and the process of predicting future citations and 2) designing two novel temporal attention mechanisms to improve the model's ability to modulate complicated temporal dependencies and to allow the model to dynamically combine the observation and prediction sides during the learning process. For the fourth of these methods, the previous work treats the traffic sensor readings as the features and views the incident duration prediction as a feature-driven regression, which typically suffers from three drawbacks: 1) ignoring the existence of the road-sensor hierarchical structure in the real-world traffic network, 2) unable to learn and modulate the hidden temporal patterns in the sensor readings, and 3) lack of consideration of the spatial connectivity between arterial roads and traffic sensors. This work makes three significant contributions: 1) designing a hierarchical graph convolutional network architecture for modeling the road-sensor hierarchy, 2) proposing novel spatiotemporal attention mechanism on the sensor- and road-level features for representation learning, and 3) presenting a graph convolutional network-based method for incident representation learning via spatial connectivity modeling and traffic characteristics modulation. / Doctor of Philosophy / Information mining and knowledge learning from sequential data is a field of growing importance in both industrial and academic fields. Sequential data, which is the natural representation format of the information flow in many applications, usually carries enormous information and is able to help researchers gain insights for various tasks such as airport threat detection, cyber-attack detection, recommender system, point-of-interest (POI) prediction, and citation forecasting. This dissertation focuses on developing the methods for sequential data-driven applications and evolutionary dynamics characterization for various topics such as transit service disruption detection, early event detection on social media, technology opportunity discovery, and traffic incident impact analysis. In particular, four specific applications are studied with four proposed novel methods, including a spatiotemporal feature learning framework for transit service disruption detection, a multi-task learning framework for cybersecurity event detection, citation dynamics modeling via multi-context attentional recurrent neural networks, and traffic incident impact forecasting via hierarchical spatiotemporal graph neural networks. For the first of these methods, the existing transit service disruption detection methods usually suffer from two significant shortcomings: 1) failing to modulate the sparsity of the social media feature domain, i.e., only a few important ``particles'' are indeed related to service disruption among the massive volume of data generated every day and 2) ignoring the real-world geographical connections of transit networks as well as the semantic consistency existing in the problem space. This work makes three contributions: 1) developing a spatiotemporal learning framework for metro disruption detection using open-source data, 2) modeling semantic similarity and spatial connectivity among metro lines in feature space, and 3) developing an optimization algorithm for solving the multi-convex and non-smooth objective function efficiently. For the second of these methods, the conventional studies in cybersecurity detection suffer from the following shortcomings: 1) unable to capture weak signals generated by the cyber-attacks on small organizations or individual accounts, 2) lack of generalization of distinct types of security incidents, and 3) failing to consider the relatedness across different types of cyber-attacks in the feature domain. Three contributions are made in this work: 1) formulating the problem of social media-based cyber-attack detection into the multi-task learning framework, 2) modeling multi-type task relatedness in feature space, and 3) developing an efficient algorithm to solve the non-smooth model with inequality constraints. For the third of these methods, conventional citation forecasting methods are using the traditional temporal point process, which suffers from several drawbacks: 1) unable to predict the technological categories of citing documents and thus are incapable of technological diversity assessment, and 2) require prior domain knowledge and thus are hard to extend to different research areas. Two contributions are made in this work: 1) formulating a novel framework to provide long-term citation predictions in an end-to-end fashion by integrating the process of learning intensity function representations and the process of predicting future citations and 2) designing two novel temporal attention mechanisms to improve the model's ability to modulate complicated temporal dependencies and to allow the model to dynamically combine the observation and prediction sides during the learning process. For the fourth of these methods, the previous work treats the traffic sensor readings as the features and views the incident duration prediction as a feature-driven regression, which typically suffers from three drawbacks: 1) ignoring the existence of the road-sensor hierarchical structure in the real-world traffic network, 2) unable to learn and modulate the hidden temporal patterns in the sensor readings, and 3) lack of consideration of the spatial connectivity between arterial roads and traffic sensors. This work makes three significant contributions: 1) designing a hierarchical graph convolutional network architecture for modeling the road-sensor hierarchy, 2) proposing novel spatiotemporal attention mechanism on the sensor- and road-level features for representation learning, and 3) presenting a graph convolutional network-based method for incident representation learning via spatial connectivity modeling and traffic characteristics modulation.

This thesis consists of three manuscripts, which are presented in chapters 2, 3 and 4. The first manuscript (chapter 2) is a protocol for systematic review which outlines the steps used to systematically analyze the literature with regards to the primary objective of reviewing available evidence. The second manuscript (chapter 3) is the presentation of systematic review and meta-analysis in the form of a manuscript to be submitted for peer-review by fall 2020. The third manuscript (chapter 4) summarizes an analysis of CHILD study data that assesses the association of dietary patterns with infant wheeze and asthma through age 3. In chapter 5, the thesis concludes by summarizing study limitations, epidemiological implications, clinical relevance of these findings and future directions. Appendices are also included to highlight research methodology and describe comprehensively all data analyses. / Thesis / Master of Science (MSc) / This paper-based thesis includes my graduate research work to satisfy the requirements for a masters in science (M.Sc.) degree in the Medical Sciences department. The focus of this thesis is to contribute an important study of the role of maternal and infant nutrition in the development of asthma and wheeze in children. My research findings are presented as original manuscripts of a systematic review protocol, systematic review, and a primary analysis of data collected as part of the CHILD longitudinal birth cohort study. At present, the first 2 manuscripts are under review by scientific journals; and the third will be submitted for peer-review in October 2021.

Asthma is a chronic disease of airway hyper-responsiveness, airway inflammation and episodic bronchoconstriction. With asthma forecasted to increase by an additional 100 million cases by 2025, there is a critical and immediate need to address new asthma therapies. Guidelines for asthma treatment in the emergency department conditionally recommend intravenous magnesium sulfate (MgSO4). However, some investigations have failed to demonstrate beneficial effects. Ethical constraints limit evaluation of the bronchodilatory effects of MgSO4 alone in patients with acute asthma exacerbation, independent of other conventional therapeutics. To address this ethical dilemma, this study consisted of two phases: 1) quantification of the independent pulmonary effect of three doubling doses of MgSO4 in the spontaneous equine model of asthma during naturally occurring exacerbations of bronchoconstriction, and 2) evaluation of arterial blood gas parameters in response to administration of MgSO4 at a dose identified in phase 1 that yielded greatest efficacy without deleterious side effects.

As a complicated lethal medical emergency, sepsis is not easy to be diagnosed until it is too late for taking any life saving actions. Early prediction of sepsis in ICUs may reduce inpatient mortality rate. Although deep learning models can make predictions on the outcome of ICU stays with high accuracies, the opacity of such neural networks decreases their reliability. Particularly, in the ICU settings where the time is not on doctors' side and every single mistake increase the chances of patient's mortality. Therefore, it is crucial for the predictive model to provide some sort of reasoning in addition to the prediction it provides, so that the medical staff could avoid actions based on false alarms. To address this problem, we propose to add an attention layer to a deep recurrent neural network that can learn the relative importance of each of the parameters of the multivariate data of the ICU stay. Our approach sheds light on providing explainability through attention mechanism. We compare our method with some of the state-of-the-art methods and show the superiority of our approach in terms of providing explanations.

Pulmonary surfactant is synthesized in the alveoli and lines the respiratory epithelium of the airways. Phospholipids, the main component of surfactant, confer it its ability to lower surface tension and to prevent alveolar collapse. Airway surfactant helps maintain smaller airway patency, improves muco-ciliary clearance, decreases bronchoconstriction, and modulates pulmonary immunity. Surfactant alterations in human asthma are therefore believed to contribute to the severity of airway obstruction. The goal of our first study was to characterize surfactant phospholipid composition and function in healthy horses, and to investigate the influence of age and bronchoalveolar lavage fluid (BALF) sample characteristics on surfactant. For that purpose, BALF was collected from 17 healthy horses and evaluated for BALF recovery percentage, cell count, and cell differential. BALF was separated into crude surfactant pellets (CSP) and supernatant and was analyzed for phospholipid content, protein content, phospholipid composition, and surface tension. Interestingly, phospholipid (surfactant) content in CSP significantly decreased with age. BALF recovery percentage, nucleated cell count, and cytological profile did not affect surfactant composition or function. The hypothesis of our second study was that surfactant alterations in RAO-affected horses are related to clinical stage of RAO. The objectives were 1) to compare surfactant phospholipid composition and function between Non-RAO and RAO horses at clinical stages and 2) to investigate relationships between surfactant alterations and variables assessing clinical stage of RAO. Seven horses with confirmed RAO and seven Non-RAO horses were evaluated in pairs (RAO/Non-RAO) at baseline, during exposure to hay, and post-exposure. Assessments included: clinical scoring, measure of maximal change in pleural pressure (ΔPplmax), airway endoscopy, and BALF cell counts and differentials. Samples were processed and analyzed as described above. Phospholipid levels in BALF were significantly lower in RAO versus Non-RAO horses, even in the absence of clinical signs. In the group of RAO horses, phospholipid content was significantly lower during exposure versus baseline. Furthermore, exposure to hay led to an increase in the protein versus phospholipid ratio in BALF from RAO horses. No significant differences were found in BALF protein content, phospholipid composition, or surface tension between or within groups of horses. Phosphatidylglycerol percentage had a tendency to be lower in RAO horses with higher clinical scores. Supernatant protein content was related to BALF neutrophilia in RAO crisis and overall ΔPplmax . In conclusion, our study demonstrated that surfactant alterations in RAO horses are present in remission and are exacerbated following exposure to hay. It is conceivable that a lower amount of surfactant in bronchioli of RAO horses may contribute to the horses' propensity to develop airway obstruction, mucous accumulation, and bronchial hyperresponsiveness. This may be exacerbated during crisis by a relatively higher protein versus phospholipid ratio. Furthermore, a progressive decrease of surfactant levels in older horses may contribute to a worsening of clinical signs in older RAO-affected horses. / Ph. D.

This study is about prediction of the stockmarket through a comparison of neural networks and statistical models. The study aims to improve the accuracy of stock prediction. Much of the research made on predicting shares deals with statistical models, but also neural networks and then mainly the types RNN and CNN. No research has been done on how these neural networks can be combined, which is why this study aims for this. Tests are made on statistical models, neural networks and combined neural networks to predict stocks at minute level. The result shows that a combination of two neural networks of type RNN gives the best accuracy in the prediction of shares. The accuracy of the predictions increases further if these combined neural networks are trained to predict different time horizons. In addition to tests for accuracy, simulations have also been made which also confirm that there is some possibility to predict shares. Two combined RNNs gave best results, but in the simulations, even CNN made good predictions. One conclusion can be drawn that the stock market is not entirely effective as some opportunity to predict future values exists. Another conclusion is that neural networks are better than statistical models to predict stocks if the neural networks are combined and are of type RNN. / Denna studie behandlar prediktion av aktier genom en jämförelse av neurala nätverk och statistiska modeller. Studien syftar till att förbättra noggrannheten för aktieprediktion. Mycket av den forskning som gjorts om att förutspå aktier behandlar statistiska modeller, men även neurala nätverk och då främst typerna RNN och CNN. Ingen forskning har dock gjorts på hur dessa neurala nätverk kan kombineras, varför denna studie syftar till just detta. Tester är gjorda på statistiska modeller, neurala nätverk och kombinerade neurala nätverk för att förutspå aktier på minutnivå. Resultatet visar att en kombination av två neurala nätverk av typen RNN ger bäst noggrannhet vid prediktion av aktier. Noggrannheten i prediktionerna ökar ytterligare om dessa neurala nätverk tränas för att förutspå olika tidshorisont. Utöver tester för prediktionernas noggrannhet har även simuleringar genomförts som även de bekräftar att viss möjlighet finns att förutspå aktier. Två kombinerade RNN gav bra resultat, men här visade även CNN bra prediktioner. En slutsats kan dras om att aktiemarknaden inte är helt effektiv då viss möjlighet att förutspå framtida värden finns. Ytterligare en slutsats är att neurala nätverk är bättre än statistiska modeller till att förutspå aktier om de neurala nätverken kombineras och är av typen RNN.

Neste trabalho apresentamos uma versão do Teorema de Poincaré-Bendixson para campos vetoriais contínuos na garrafa de Klein. Como conseqüência, mostramos que a garrafa de Klein não possui campo vetorial contínuo com trajetória injetiva recorrente / We present a version of the Poincaré-Bendixson Theorem on the Klein bottle for continuous vector fields. As a consequence, we obtain the fact that the Klein bottle does not admit continuous vector fields having a recurrent injective trajectory

Anhand von 2019 Schilddrüsenoperationen, die in den Jahren 1985-1996 im Universitätsklinikum Charite, Standort Rudolf-Virchow-Klinikum in Berlin durchgeführt wurden, wird die Bedeutung der Operationstechnik und der Operationsindikation für die chirurgische Komplikationsrate untersucht. Insgesamt wurden 3471 Schilddrüsenlappen operiert. Anhand einer Nachuntersuchung und der direkten postoperativen Dokumentation ließen sich die postoperativen Komplikationen dokumentieren. Die Rate an transienten Rekurrensparesen betrug 4,5%, bei 0,7 der Operierten fanden sich permanente Paresen. Bezogen auf die "nerves at risk" fanden sich in 2,8% transiente und in 0,5% permanente Paresen. Die Darstellung des Nervus laryngeus rekurrens erwies sich in dieser Untersuchung als signifikant komplikationsärmer im Bezug auf permanente Paresen. Hier waren bei der Darstellung des Nervens 0,5% permanente Läsionen aufgetreten, im Gegensatz zu 0,9% permanenter Läsionen ohne Darstellung des Nerven. Der direkte Vergleich von Komplikationen der Hemithyreoidektomien und der kontralateralen subtotalen Resektion ergab keinen signifikanten Unterschied betreffs der Paresen. Die Hemithyreoidektomie mit kontralateraler subtotaler Resektion war im Vergleich zu der Thyreoidektomie oder der subtotalen Resektion beidseits die komplikationsärmste Operationsmöglichkeit. Eine postoperative Erniedrigung des Serumkalziumspiegels wurde bei 18,9% der Patienten festgestellt, aber nur 0,9% waren persistent. Die schilddrüsennahe Ligatur der Arteria thyroidea inferior ergab eine niedrigere Rate an postoperativen Hypokalzämien, die jedoch statistisch nicht signifikant war. Da aber weder eine erhöhte Rate an Rekurrensparesen, noch eine vermehrte Rezidivneigung zu befürchten ist, scheint die schilddrüsennahe Ligatur vorteilhafter zu sein. Unter 149 Rezidivoperationen fanden sich 126 "echte" Rezidive, die Rate an permanenten Rekurrensparesen betrug hier 1,6%, bezogen auf "nerves at risk". Die Rate an Hypokalzämien lag bei 23,8%, wovon 0,8% permanenter Natur waren. Patienten über 70 Jahre haben sowohl perioperativ, als auch postoperativ kein erhöhtes Risiko eine Komplikation zu erleiden, so daß Operationen an der Schilddrüse durchaus auch im hohen Alter gerechtfertigt werden können. / Abstract During the years from 1986 until 1996 2019 thyroid gland surgeries have been performed at the university clinic charite', campus virchow-clinic. That means 3471 thyroid gland lobes have been treated. This paper is investigating the complications directly caused by the surgical procedures. Data gained directly during surgery and 6 month after surgery allow a very detailed view on the complications. There have been 4.5% transient recurrent nerve palsies and 0.7% permanent. Data based on the "nerves at risk" show 2.8% transient and 0.5% permanent palsies. A direct comparison of hemithyroidectomy and near-total thyroidectomy on the contralateral side show no significant difference in permanent nerve palsies. The hemithyroidectomy with contralateral near-total thyroidectomy was the surgical procedure with the lowest rate of complications compared to the thyroidectomy or the near-total thyroidectomy of both lobes. Postoperative Hypocalcemia was seen in 18,9% of all patients, but only 0.9% suffered from permanent hypocalcemia. The ligature of the lower thyroid artery close to the thyroid gland showed lower rates of hypocalcemia , but the results haven't been statistically significant. There has not been a higher rate of nerve palsies or a higher rate of relapses, so the ligature close to the gland is recommended. 149 operations were necessary due to recurrent growth of the thyroid gland. There have been126 real relapses, the rate of permanent nerve palsy was 1,6% for the "nerves at risk". Hypocalcemia was found in 23,8% of all cases, 0.8% were permanent. Patients older than 70 years do not have a higher risks to suffer from complications than younger patients, so thyroid surgery should also be performed for the elderly.

La détermination de la structure tertiaire du ribosome fut une étape importante dans la compréhension du mécanisme de la synthèse des protéines. Par contre, l’élucidation de la structure du ribosome comme tel ne permet pas une compréhension de sa fonction. Pour mieux comprendre la nature des relations entre la structure et la fonction du ribosome, sa structure doit être étudiée de manière systématique. Au cours des dernières années, nous avons entrepris une démarche systématique afin d’identifier et de caractériser de nouveaux motifs structuraux qui existent dans la structure du ribosome et d’autres molécules contenant de l’ARN. L’analyse de plusieurs exemples d’empaquetage de deux hélices d’ARN dans la structure du ribosome nous a permis d’identifier un nouveau motif structural, nommé « G-ribo ». Dans ce motif, l’interaction d’une guanosine dans une hélice avec le ribose d’un nucléotide d’une autre hélice donne naissance à un réseau d’interactions complexes entre les nucléotides voisins. Le motif G-ribo est retrouvé à 8 endroits dans la structure du ribosome. La structure du G-ribo possède certaines particularités qui lui permettent de favoriser la formation d’un certain type de pseudo-nœuds dans le ribosome. L’analyse systématique de la structure du ribosome et de la ARNase P a permis d’identifier un autre motif structural, nommé « DTJ » ou « Double-Twist Joint motif ». Ce motif est formé de trois courtes hélices qui s’empilent l’une sur l’autre. Dans la zone de contact entre chaque paire d’hélices, deux paires de bases consécutives sont surenroulées par rapport à deux paires de bases consécutives retrouvées dans l’ARN de forme A. Un nucléotide d’une paire de bases est toujours connecté directement à un nucléotide de la paire de bases surenroulée, tandis que les nucléotides opposés sont connectés par un ou plusieurs nucléotides non appariés. L’introduction d’un surenroulement entre deux paires de bases consécutives brise l’empilement entre les nucléotides et déstabilise l’hélice d’ARN. Dans le motif DTJ, les nucléotides non appariés qui lient les deux paires de bases surenroulées interagissent avec une des trois hélices qui forment le motif, offrant ainsi une stratégie élégante de stabilisation de l’arrangement. Pour déterminer les contraintes de séquences imposées sur la structure tertiaire d’un motif récurrent dans le ribosome, nous avons développé une nouvelle approche expérimentale. Nous avons introduit des librairies combinatoires de certains nucléotides retrouvés dans des motifs particuliers du ribosome. Suite à l’analyse des séquences alternatives sélectionnées in vivo pour différents représentants d’un motif, nous avons été en mesure d’identifier les contraintes responsables de l’intégrité d’un motif et celles responsables d’interactions avec les éléments qui forment le contexte structural du motif. Les résultats présentés dans cette thèse élargissent considérablement notre compréhension des principes de formation de la structure d’ARN et apportent une nouvelle façon d’identifier et de caractériser de nouveaux motifs structuraux d’ARN. / Although determination of the ribosome tertiary structure has been an outstanding step towards elucidation of the mechanism of protein synthesis, the complexity of this structure does not provide an easy answer of how this large molecular complex works. In order to understand the nature of structure-function relationships in the ribosome, the ribosome structure itself should be subjected to thorough analysis. In the last years, we undertook systematic efforts toward identification and characterization of all recurrent structural motifs existing in the ribosomal RNA and in other RNA-containing molecules. The analysis of many instances of helix-helix packing in the ribosome structure allowed us to identify a new structural motif which we called “G-ribo”. In this motif, an interaction of the sugar edge of a guanosine in one helix with the ribose of a nucleotide from another helix was found to be at the origin of a complex network of concomitant inter-nucleotide interactions. In total, the G-ribo motif was found at eight locations within the ribosomal RNA. A surprising feature of this motif consists in its ability to favor the formation of pseudoknots of a particular type. In the ribosome structure, there are four pseudoknots whose formation is mediated by the G-ribo motif. Systematic analysis of the ribosome as well as the RNAseP crystal structures allowed for the identification of a new RNA motif, which we called “DTJ”, or Double-Twist Joint motif. This motif is made of three short RNA double helices, which stack one on top of another. In the contact zone of each pair of helices two consecutive base pairs are over-twisted compared to the regular helical twist of 32° of A-RNA. One nucleotide of the base pair is always directly connected to the one nucleotide of the over-twisted base pair, while the opposite nucleotides of these base pairs are connected with one or several unpaired nucleotides. Introduction of the helical over-twist between two consecutive base pairs breaks the inter-nucleotide stacking and destabilizes the RNA double helix. In the DTJ, the unpaired nucleotides that connect the two over-twisted base pairs interact with one of the three motif-forming helices, providing an elegant strategy for the stabilization of the whole arrangement. To determine the nucleotide sequence constraints imposed on the structure of recurrent RNA motifs in the functional ribosome we developed a new approach consisting in the selection of functional ribosomes from a combinatorial gene library in which certain nucleotides of the rRNA gene corresponding to a particular motif were randomized. Comparison of the constraints determined for different examples of the same motif allowed us to distinguish between constraints responsible for the integrity of the motif and for its interaction with surrounding elements, including ribosomal proteins. The work significantly improves our understanding of the principles of RNA structure formation and opens a new way to identify and characterize RNA motifs.