22455 research outputs found
Sort by
Influence of hypoxia and vascularization on the osteogenic differentiation and mineralization capacity of primary osteoporotic human mesenchymal stem cells
Get PDFThe treatment of osteoporotic fractures that are associated with critical bone defects or non-unions is a significant clinical problem. To date, autografts have served as the gold standard for bone grafts. However, autografts have been associated with many risks. Bone tissue engineering (BTE) represents a potential alternative for the treatment of these complicated fractures. Although BTE research has largely improved, the translation into clinical applications has yet to be approved by regulatory agencies. As several companies have received approval for the use of intra-operation cell isolation systems, inadequate vascularization of scaffolds has been identified as a major problem. An additional concern is the reduced osteogenic differentiation potential of human mesenchymal stem cells (hMSCs) in elderly patients. Therefore, many research groups have examined how to improve vascularization and osteogenic differentiation potential. One widespread approach for manufacturing prevascularized scaffolds has been a co-culture of scaffolds with hMSCs and human umbilical vein endothelial cells (hUVECs) combined with the testing of different oxygen levels in cell culture. The aim of the present study was therefore to gain knowledge about the influence of oxygen on the proliferation,growth characteristics, and osteogenic differentiation of hMSCs from different donors, specifically patients with high-energy fracture trauma (HET) and osteoporotic fracture trauma (OFT). Furthermore, co-cultures from donors of the HET group and OFT group with hUVEs, undergoing osteogenic differentiation were analyzed. For this study, hMSCs were isolated from different donor sites and randomized into the two donor groups HET or OFT. Initially, the morphology, growth characteristics, proliferation (CumPD and colony forming units (CFU)) and osteogenic differentiation potential of hMSCs in normoxia and hypoxia were analyzed. Further, the osteogenic differentiation potential of co-cultures (hMSCs:hUVECs) in different ratios was tested in 2D and 3D.
Analysis of the morphology aspect in hypoxia and normoxia showed a tendency towards flattening of the cells for both groups (HET and OFT) in normoxia. Following cumulative population doubling in normoxia versus hypoxia, both groups showed a non-significant tendency toward improved proliferation in hypoxia. Further osteogenic differentiation was examined, showing that osteogenic differentiation in hypoxia was non-significantly reduced for both groups (HET and OFT). Comparison of the two groups, HET and OFT, found that the proliferation capacity, CFU capacity, and osteogenic differentiation potential of the HET group had a tendency towards better performance, although that difference was not always significant. Further co-cultures (hMSCs:hUVECs) of different ratios (1:0, 1:2, 1:3) were analyzed to gain a better understanding of the interaction between the cell types and the osteogenic differentiation potential in co-cultures. Most donor samples showed a tendency towards a higher osteogenic differentiation capacity in co-cultures than in hMSCs monocultures. The influence of the co-culture was more pronounced for OFT hMSCs than for the HET hMSCs.
Although many results were non-significant, a tendency towards improved proliferation and later senescence in hypoxia, stronger osteogenic differentiation capacity in normoxia and a better osteogenic differentiation capacity in co-cultures demonstrates that hMSCs are very sensitive to oxygen levels and their behavior is influenced by cocultures. The knowledge of the specified pathways behind these interactions and the best conditions for hMSCs and hUVECs still have to be investigated. All these findings will help identify a better solution for non-union fractures and improve BTE
Machine learning approaches to latent variable modeling
Get PDFDiese Arbeit enthält vier Beiträge (Manuskripte I bis IV), die jeweils neue methodische Ansätze zum Umgang mit Verzerrungen und Bias in mehrdimensionalen IRT-Modellen einführen. Insbesondere wird das Potenzial nichtparametrischer, maschineller Lernverfahren eingehend untersucht. Die im Rahmen dieser Arbeit verfassten Manuskripte stellen Methoden zur Schätzung von Modellparametern und latenten Variablen-Scores multidimensionaler IRT-Modelle vor. Diese Methoden berücksichtigen die Verzerrung, die ungemessene und/oder gemessene Kovariaten auf die Parameterschätzung haben können.
In Manuskript I wird gezeigt, dass die Einbeziehung von latenten Item-Effekt-Variablen in longitudinale IRT-Modelle für ordinale Antwortvariablen interindividuelle Unterschiede in den Item-Schwierigkeits-Parametern kontrollieren kann. Auf diese Weise wird die Verzerrung, die gemessene oder nicht gemessene Kovariaten auf die Schätzung der Item-Schwierigkeits-Parameter haben können, berücksichtigt.
Außerhalb der Längsschnittforschung ist es nicht möglich, solche Item-Effekt-Variablen zu schätzen. Interindividuelle Unterschiede in den Item-Parametern, die auch als Differential Item Functioning (DIF) bezeichnet werden, können jedoch mit Hilfe von Model Based Recursive Partitioning (MOB) berücksichtigt werden, einem algorithmischen Modellierungsansatz, der aus den Methoden des maschinellen Lernens stammt. Manuskript II zeigt, dass MOB zur Kontrolle von DIF in mehrdimensionalen IRT-Modellen verwendet werden kann. Dies funktioniert, indem automatisch Untergruppen mit stabilen Item-Parameterschätzungen erkannt werden.
Manuskript III stellt eine Methode zur Schätzung latenter Variablen-Scores von Individuen vor, die in Bezug auf bestimmte gemessene Kovariaten unverzerrt sind. Zu diesem Zweck wird ein Ensemble von MOB-Trees gebildet. Innerhalb des MOB-Tree-Ensembles werden Untergruppen mit stabilen Item-Parameter-Schätzungen verwendet, um latente Variablen-Scores zu schätzen, die in Bezug auf relevante Untergruppen in der Population unverzerrt sind. Somit sind diese latenten Variablen-Scores im Hinblick auf systematische Einflüsse dieser gemessenen Kovariablen interpretierbar, ohne durch diese Variablen verzerrt zu werden.
Um einen MOB-Tree zu erstellen, muss ein Parameterinstabilitätstest wiederholt für ein (mehrdimensionales) IRT-Modell berechnet werden. Mehrdimensionale IRT-Modelle werden effizient als ordinale Faktorenmodelle geschätzt. Für das Modell wird die erste Ableitung der Zielfunktion (d.h. die Score-Funktion) verwendet, um die Parameterinstabilität zu schätzen. In Manuskript IV wird daher eine Methode zur Schätzung der individuellen Beiträge zu dieser Funktion für ordinale Faktorenmodelle vorgeschlagen. Dadurch wird es möglich, viele Parameterinstabilitätstests für mehrdimensionale IRT-Modelle in kurzer Zeit zu berechnen.
Die mit diesen vier Beiträgen vorgestellten Methoden ermöglichen die effiziente Berechnung von Parameterinstabilitätstests für mehrdimensionale IRT Modelle, die Schätzung individueller Schwierigkeits-Parameter in Längsschnittkontexten und latenter Variablen-Scores, die außerhalb von Längsschnittkontexten in Bezug auf spezifische gemessene Kovariaten unverzerrt sind.This thesis contains four contributions (Papers I to IV) which present approaches to dealing with bias in multidimensional IRT models. In particular, the potential of nonparametric tree-based machine learning methods is examined in detail. The papers written in the scope of this thesis provide methods to estimate model parameters and latent variable scores of multidimensional IRT models while considering the bias that unmeasured and/or measured covariates may have on parameter estimation.
In Paper I, it is shown that the inclusion of latent item effect variables in longitudinal IRT models for ordinal response variables can control for inter-individual differences in item difficulty parameters. This way, the bias that measured or unmeasured covariates may have on the estimation of the item difficulty parameters is taken into account.
Outside of longitudinal research, it is not possible to estimate such item effect variables. However, inter-individual differences in item parameters, also referred to as Differential Item Functioning (DIF), can be accounted for via Model Based Recursive Partitioning (MOB), an algorithmic modeling approach borrowed from the tree-based methods of machine learning. Paper II illustrates that MOB can be used to control for DIF in multidimensional IRT models. For such models, MOB may be used to automatically detect subgroups with stable item parameter estimates.
Paper III introduces a method to estimate latent variable scores of individuals that are unbiased with respect to certain measured covariates. For this, an ensemble of MOB trees is grown. Within the MOB tree ensemble, subgroups with stable item parameter estimates are used to estimate latent variable scores that are unbiased with respect to relevant subgroups in the population. Thus, these latent variable scores are interpretable with respect to systematic influences of specific measured covariates without being biased by these variables.
In order to grow a MOB tree, a parameter instability test must be computed repeatedly for a fitted (multidimensional) IRT model. Multidimensional IRT models are efficiently fitted as ordinal factor models. For the fitted model, the first derivative of the objective function (i.e.~the score function) is used to estimate parameter instability. In Paper IV, a method for the estimation of individual contributions to this score function for ordinal factor models is therefore proposed. This makes it computationally feasible to repeatedly compute parameter instability tests for multidimensional IRT models.
The methods introduced with these four contributions make it possible to efficiently compute parameter instability tests for MIRT models, to estimate individual difficulty parameters in longitudinal settings and latent variable scores that are unbiased w.r.t.~specific measured covariates outside of longitudinal settings
Erhöhte Produktion reaktiver Sauerstoffspezies und mitochondriale Stoffwechselverschiebungen in CD4+ T-Zellen von Pferden mit equiner rezidivierender Uveitis
Get PDFEtablierung und Validierung eines minimalinvasiven Kaninchenmodells atherosklerotischer Erkrankungen
Get PDF
Characterization of novel Modified Vaccinia virus Ankara (MVA)-based vaccines targeting hemagglutinin proteins from influenza A virus subtypes H2N2, H2N9 and H5N8
Get PDFThermosensitives Liposomales Doxorubicin (DPPG₂-TSL-DOX) simultan mit regionaler Hyperthermie als initiale Therapie von inoperablen Weichteilsarkomen
Get PDFSzegő-type asymptotics for the free Dirac operator
Get PDFDie Resultate der vorliegenden Dissertation sind maßgeblich durch die Untersuchung von Skalierungsgesetzen der Verschränkungsentropie für freie relativistische Fermionen motiviert. Diese stehen im Zusammenhang mit Szegő-Asymptotiken für Spektralprojektionen des freien Dirac-Operators, wobei als Testfunktion eine Rényi-Entropiefunktion gewählt wird. Solch eine Spektralprojektion kann als Integraloperator mit einem, möglicherweise unstetigen, matrixwertigen Symbol aufgefasst werden. Im Falle unstetiger skalarwertiger Symbole ist die Szegő-Asymptotik Inhalt der Widom–Sobolev Formel. Eine Anwendung dieser Formel ergibt den Beweis eines logarithmisch verstärkten Oberflächengesetzes, also einem führenden Term der Ordnung L^(d−1) log L bezüglich des Skalierungsparameters L in der Asymptotik, im nicht relativistischen Fall des freien d-dimensionalen Schrödinger-Operators. Die logarithmische Verstärkung tritt auf, falls das Abschneiden an der Fermi-Energie innerhalb des absolutstetigen Spektrums des freien Schrödinger-Operators liegt. Im Fall einer nicht-positiven Fermi-Energie ist das entsprechende Symbol hingegen effektiv glatt und es tritt höchstens ein Oberflächengesetz, also ein führender Term der Ordnung L^(d−1), auf.
Das erste Resultat dieser Dissertation ist in einer Zusammenarbeit mit Peter Müller entstanden und ist eine Verallgemeinerung der Widom–Sobolev Formel auf matrixwertige Symbole, welche unstetig auf dem (d−1)-dimensionalen Rand eines hinreichend regulären Bereiches sind. Es werden drei, in ihrer Allgemeinheit aufsteigende, Klassen von Testfunktionen betrachtet. Die allgemeinste dieser Klassen enthält die Rényi-Entropiefunktionen. Mit zunehmender Allgemeinheit der Testfunktionen sind striktere Voraussetzungen an die Klasse der zulässigen Symbole verbunden. Auch wenn es für den Anwendungsfall des freien Dirac-Operators nicht nötig ist, werden keine Annahmen an die Kommutativitätseigenschaften des matrixwertigen Symbols benötigt. Der Koeffizient des resultierenden verstärkten Oberflächengesetzes ist genauso explizit wie im skalaren Fall. Dies steht im Gegensatz zur Situation bei glatten Symbolen. Hier ist der Koeffizient des zugehörigen Oberflächengesetzes deutlich weniger explizit für matrixwertige Symbole als für skalarwertige Symbole.
Das nächste Resultat, ebenfalls basierend auf der Zusammenarbeit mit Peter Müller, ist eine Anwendung der bewiesenen Widom–Sobolev Formel für matrixwertige Symbole auf den Spezialfall des freien Dirac-Operators. Da das Spektrum des Dirac-Operators für negative Energien unbeschränkt ist, betrachten wir eine glatt abgeschnittene Version der Fermi-Projektion, um zu garantieren, dass der betrachtete Operator Spurklasse ist. Das Symbol der Projektion erfüllt, abhängig von den beiden Parametern Masse und Fermi-Energie, unterschiedliche Eigenschaften. Wenn, in beliebiger Dimension, der Absolutbetrag der Fermi-Energie strikt größer als die Masse ist oder Fermi-Energie und Masse im eindimensionalen Fall verschwinden, weist das Symbol eine (d−1)-dimensionale Unstetigkeit auf und es gilt ein verstärktes Oberflächengesetz. Der dazugehörige Koeffizient ist unabhängig vom glatten Abschneiden der Fermi-Projektion. In den anderen Fällen wird gezeigt, dass höchstens ein Oberflächengesetz auftreten kann.
Ein besonderer Fall tritt auf, falls Fermi-Energie und Masse in einem mindestens zweidimensionalen System verschwinden. In diesem Fall weist das Symbol aufgrund der Struktur des freien Dirac-Operators eine Unstetigkeit in einem einzigen Punkt auf, eine Situation die im nicht relativistischen Fall nicht auftritt. Da diese Unstetigkeit nicht hinreichend für ein verstärktes Oberflächengesetz ist, werden stattdessen die Terme niedrigerer Ordnung der Asymptotik betrachtet. Dazu erfolgt eine Einschränkung auf Würfel als Abschneidebereiche im Ort und analytische Testfunktionen. Das letzte Resultat dieser Dissertation zeigt, dass sich die asymptotische Entwicklung ab dem (d+1)ten Term von der Entwicklung für glatte Symbole unterscheidet. Die ersten d Terme der Asymptotik werden bestimmt und es wird bewiesen, dass der übrigbleibende Fehler von logarithmischer Ordnung, log L, ist. Im Spezialfall, dass die Testfunktion ein Polynom von Grad drei oder niedriger ist, wird eine Entwicklung mit d+1 Termen bewiesen, wobei der zusätzliche Term von logarithmischer Ordnung und der Fehlerterm von konstanter Ordnung ist. Der Koeffizient des logarithmischen Terms ist unabhängig vom glatten Abschneiden der Fermi-Projektion. Die Strategie dieses Beweises beruht auf der Tatsache, dass die inverse Fourier-Transformation des Symbols homogen vom Grad −d ist.The primary motivation behind the results presented in this thesis is the study of scaling laws for the entanglement entropy of free relativistic fermions. It is closely related to the Szegő-type asymptotics for spectral projections of the free Dirac operator with the test function given by a Rényi entropy function. Such a projection can be written as an integral operator with, potentially discontinuous, matrix-valued symbol. The study of Szegő-type asymptotics for scalar-valued discontinuous symbols is the subject of the Widom–Sobolev formula. As a consequence of this formula, a rigorous proof of a logarithmically enhanced are law, i.e. a scaling of leading order L^(d−1) log L in the scaling parameter L, of the entanglement entropy has been obtained in the non-relativistic case of the free d-dimensional Schrödinger operator. The logarithmic enhancement occurs if the cut-off at the Fermi energy is inside the absolutely continuous spectrum of the free Schrödinger operator. In the case of a non-positive Fermi energy, the symbol of the corresponding pseudo-differential operator is effectively smooth, yielding at most an area law, i.e. a scaling of leading order L^(d−1).
In the first result of this thesis, based on joint work with Peter Müller, we extend the Widom–Sobolev formula for scalar-valued symbols to matrix-valued symbols which are discontinuous at the (d−1)-dimensional boundary of a suitable domain. We consider three different classes of test functions, increasing in generality. The most general of these classes of test functions contains the Rényi entropy functions. As the test functions increase in generality we require more restrictive assumptions on the class of symbols. We do not require any assumptions on the commutation properties of the matrix-valued symbol. The coefficient of the obtained enhanced area term is as explicit as in the scalar-valued case. This is in contrast to the case of a smooth symbol, where the coefficient of the corresponding area law is substantially less explicit in the matrix-valued case.
In the next result, also based on joint work with Peter Müller, we apply the obtained Widom–Sobolev formula for matrix-valued symbols to the special case of the free Dirac operator. Due to the Dirac Sea being unbounded at negative energy levels, we consider a smoothly truncated version of the Fermi projection in order to guarantee that the operator in question is trace class. We analyse the resulting symbol and distinguish between several cases, depending on both mass and Fermi energy. If, in arbitrary dimension, the modulus of the Fermi Energy is strictly larger than the mass, or we have Fermi energy zero in the one-dimensional massless case, the symbol features a suitable (d−1)-dimensional discontinuity and we obtain an enhanced area law with coefficient independent of the smooth truncation of the Fermi projection. In the other cases we show that at most an area law holds.
A special case occurs when both Fermi energy and mass vanish in dimension larger than one. In this case, the structure of the free Dirac operator gives rise to a symbol which is discontinuous at a single point, a situation not encountered in the non-relativistic case. As this discontinuity is not sufficient to yield a logarithmic enhancement of the area law, we study the lower-order terms of the asymptotic expansion. We restrict ourselves to nalytic test functions and cubes as spatial cut-off domains. We show that the expansion starts to differ from the expansion for smooth symbols starting from the (d+1)st term. More explicitly, we obtain the first d terms of the asymptotic expansion and prove that the error obtained by subtracting these first d terms from the expansion is of logarithmic order in the scaling parameter L, instead of being of constant order as in the case of a smooth symbol. In the special case that the test function is a polynomial of degree less or equal than three, we obtain a (d+1)-term expansion with the lowest-order term being of order log L and the error term being of constant order. The coefficient of this logarithmic term is also independent of the smooth truncation of the Fermi projection. The key to the required analysis is the fact that, in the case of vanishing mass and Fermi energy, the inverse Fourier transform of the symbol is homogeneous of degree −d
Into the stellar glare
Get PDFThe direct detection and characterisation of gas giant exoplanets is a key method for investigating the formation, evolution and atmospheric properties of these distant worlds. Yet, at present, the intense glare of the host stars and instrumental intricacies of the technique limit its application to massive planets orbiting their host stars on wide, far-out orbits.
To shed light on the formation history of our own Solar System and assess how representative it is among other stellar systems, we need to expand the sample size of well-characterised exoplanets at short orbital separations.
In this thesis, I present a novel approach to extending our direct detection sensitivity towards lower companion masses and shorter orbital separations. This effort builds on the combination of the astrometric and direct detection methods and enables a thorough and efficient characterisation of individual planets. In this sense, the work presented in this thesis lays the groundwork for building a population-level sample of directly detected, short-separation exoplanets.
The structure in which the different topics are presented is intended to convey a coherent narrative that illustrates an advance in our direct detection targeting capabilities.
This underlying theme draws a line from mostly blind direct imaging studies to semi-targeted approaches based on an informed target selection of stars likely to host an exoplanet, to ultimately arrive at a fully-targeted technique that is capable of predicting the exact position of a planet candidate relative to its host.
After a brief introduction of the topics upon which this thesis rests, I present an updated study of HIP 99770 b, the first exoplanet directly detected on the basis of long-term proper motion irregularities presented by its host star.
Drawing on data obtained by the GRAVITY interferometer, I constrain the orbital solution of the companion, infer its age and determine a set of atmospheric parameters. The results obtained from this thorough analysis offer valuable insights into the potential and limitations of the proper motion technique and will eventually help clarify the formation history of HIP 99770 b.
I next demonstrate how astrometric data collected by the Gaia space telescope can be used to reliably predict the position of substellar companions relative to their hosts. This new technique differs from the proper motion method in that it facilitates precise and efficient GRAVITY follow-up observations in a fully targeted manner. After the successful detection and confirmation of eight candidates, I show how the combination of the underlying Gaia data with a single GRAVITY astrometric epoch can enable tight constraints on the orbits and dynamical masses of the targeted companions and thereby reveal their true nature: five of the newly detected companions are shown to be substellar.
I continue by applying this technique to companion candidates in the planetary mass regime. After describing the target selection process, I present an analysis of the observational data we obtained for a single companion candidate and show that the target system actually corresponds to a stellar binary. I address how these objects can be mistaken for planet-hosting systems and discuss the non-detection in the context of other studies investigating the false-positive contamination of the Gaia data set.
Applying the same method Gaia employs to detect planets around stars, I next demonstrate how the unmatched astrometric precision enabled by optical and near-infrared interferometry can be exploited to detect moons around exoplanets. Simulating the gravitational perturbations they induce in planetary orbits, I compute the first exomoon sensitivity curves for different interferometric instruments.
To date, no such object has been robustly detected. Yet, they are likely to exist in large numbers and -- once detected -- exomoons will impact current theories of planet formation and our search for habitable worlds.
Finally, I provide an outlook at what lies ahead for the techniques developed in this thesis. From the likely implications of upcoming data releases and observational facilities on exoplanet science to the prospects for exomoon detection and characterisation in the more distant future, I discuss a set of flagship science cases that serve to further contextualise the progress made in this thesis.
Overall, the work presented in this thesis demonstrates how to harness precision astrometry for informing follow-up observations and characterising gas giant exoplanets.
It highlights the potential of leveraging synergies between different detection methods and the unique capabilities that optical and near-infrared interferometry can bring to exoplanet research.
All things considered, it paves the way towards building a population-level sample of directly detected planets resembling those we see in our own Solar System and understanding its formation history
The proteome of newborns with maternal obesity
Get PDFBackground
The prevalence of overweight including obesity has been growing rapidly over the past years. Changes in the intrauterine environment due to maternal overweight incl. obesity impact health after birth. Leading specifically to child- and adulthood obesity and a higher risk of developing cardiovascular and metabolic diseases. In this study, we compared the proteome of full-term newborns with maternal overweight and obesity during pregnancy, with the proteome of newborns with normal maternal weight using liquid chromatography-mass spectrometry. We aimed to identify affected proteins and differently regulated signaling pathways caused by the different intrauterine environment of mothers with overweight and obesity. We aimed to provide a different perspective and potentially new target for prevention programs.
Material and Methods
Between February 2017 and June 2019, we included 15 newborns with maternal obesity, defined as a maternal BMI > 30 m2/kg at the beginning of pregnancy and 344 newborns with maternal BMI < 25 m2/kg. Medical history and anthropometric data were taken from the maternal medical records as well as from a questionnaire. The blood sample was collected on a dried blood spot card and the proteomic analysis was performed by Mass Spectrometry at the Max Planck Institute. Afterwards, the proteomic and descriptive patient data were statistically analyzed.
Results
Analysis of the proteomic data showed a notably higher abundance of ribosomal protein S21 and Kallistatin in newborns with maternal obesity. Contrarily, Prostaglandin E2 receptor EP3 was less abundant in newborns with maternal obesity compared to newborns with normal maternal weight.
Conclusions
Based on this study we propose that maternal obesity during pregnancy may cause changes in certain protein levels in the newborn. Since our initial results did not hold up after FDR correction, presumably due to the small sample size, further research with a larger sample size is necessary. Additionally, a follow-up study would be interesting to see if the changes in the newborn persist maybe even up until adulthood