Dans les modules de puissance à connexion filaire de type wire-bonding (WB), les forts courants commutés (jusqu’à 200A pour une puce de 10x10mm²) imposent de faibles résistances et inductances d’interconnexion pour réduire la chute de tension et les surtensions. Pour cette raison, les concepteurs multiplient les fils de bonding de grand diamètre (jusqu’à 500μm) en parallèle. De plus, quand la surface de puce le permet, les WB sont soudés à au moins deux endroits différents pour améliorer la distribution du courant. A la différence d’un assemblage standard de type WB, dans un module de puissance de type Direct-Lead-Bonding (DLB), la puce et la diode sont généralement brasées d’un côté, via la technique du flip-chip, sur le dissipateur intégré. L’autre face est brasée ou frittée directement sur une broche (ou clip) interne large pour former la maille électrique grâce à une brasure à base d'étain, d’argent et de cuivre (SAC ou Sn-Ag-Cu), très épaisse pour éviter le claquage broche-terminaison de puce. Par conséquent, le DLB peut offrir une surface de contact plus performante sur les plans électrique et thermique que le WB, réduisant ainsi la résistance de contact d’environ 50% selon la bibliographie (d’un facteur dix selon nos simulations électromagnétiques), améliorant la distribution du courant dans les puces et homogénéisant la température au sein du composant. De plus, l’inductance parasite interne peut être atténuée de 57% comparé au WB selon la littérature. Si l’on aborde la dimension sécuritaire, la tenue en surintensité ou I²T d’un module de puissance WB rempli de gel de silicone est faible et procure un effet fusible naturel bien qu’imparfait (mode de défaut circuit-ouvert). Les fils de bonding subissent un phénomène de soulèvement même si leur design n’a pas été pensé dans ce sens. En remplaçant le gel par de la résine époxy, ce comportement se dégrade pour donner un mode de défaut intermédiaire dû à la limitation en température de la résine. A l’inverse, le DLB devrait montrer un très fort I²T et donc, un mode de défaut se rapprochant du court-circuit. Ces travaux proposent une approche innovante sur le thème du design des topologies de conversion sécurisées à tolérance de panne : pourquoi ne pas construire une topologie autour du mode de défaillance intrinsèque d’un module de puissance, au lieu de mettre en place des moyens classiques pour le contrecarrer, i.e. essayer d'isoler systématiquement le défaut avec des fusibles ? Le module de puissance DLB était le candidat idéal pour mettre à l’œuvre notre philosophie. Dans un premier temps, nous avons cherché à comparer les modes de défaillance des deux technologies, WB et DLB, grâce à des essais destructifs d’énergies maîtrisés. Les résistances de défaut, énergies critiques et I²T ont été mesurées sur un banc dédié, de même que des analyses d'endommagements des zones de défaillance ont été réalisées au sein du CNES-THALES Lab de Toulouse par une méthode non intrusive de type Lock-In-Thermography (LIT). Il a été montré que la technologie DLB pouvait présenter une résistance de défaut dix fois plus faible que celle de la technologie WB à même surface de puce et à même énergie de destruction. La présence du clip permet aussi de réduire le gradient thermique dans la région du défaut et de moins contraindre thermiquement l'encapsulant par rapport à la technologie WB. La forte épaisseur du joint de brasure broche – puce garantit aussi une meilleure métallisation par refusion de la zone de perçage et ainsi une résistance de défaut plus faible. Dans un second temps, les modules détruits WB et DLB ont subi des essais d’endurance sur 5 semaines, afin d’éprouver la robustesse et la stabilité de leur résistance de défaut à faible et fort courants. Les résultats montrent clairement la supériorité de la technologie DLB. Par la suite, une campagne de caractérisation thermique (Rth/Zth) des deux technologies WB et DLB a été réalisée sur la base d'un banc développé à cet effet. / In wire-bonding (WB) power-modules, high current commutated by fast power chips (up to 200A on a 10x10mm² chip) implies low resistance and low stray inductance interconnections in order to reduce voltage drop and overvoltage. For this purpose, designers use numerous large-diameter bonding wires (up to 500μm) in parallel. Whenever the die surface is large enough (like IGBTs), bonding wires are soldered at least in two different spots to improve current distribution. Compared to conventional WB structure, inside Direct-Lead-Bonding (DLB) power-modules, chip and diode are generally soldered on one side, using flip-chip method, to the heat spreader. The other side is directly soldered or sintered to the large inner lead (or clip) to form the electrical loop with a thick standard SAC soldering (Sn-Ag-Cu) in order to avoid electrical breakdown between chip and clip. Therefore, DLB would provide a wider bonding area than WB design, reducing the emitter contact resistance by almost 50% in the literature (by a factor of 10 according to our simulation results), improving current uniformity in the chips and thus resulting in a uniform surface temperature distribution inside the device. Besides, DLB internal inductance could be reduced to 57% of wire-bonded modules according to literature. Considering safety aspects, the overcurrent capability of a gel-filled wire-bonding power module is low and provides a natural but imperfect wire-fuse-effect (as an open-failure mode). Lift-offs happen even if WB design is not optimized for it. Replacing the gel with an epoxy resin, this behaviour gets worse and an intermediate failure-mode is reached due to the epoxy temperature limitation. On the opposite, DLB should have a very high overcurrent capability characteristic and thus short-failure mode behaviour. This work offers a quite new approach in the field of fault-tolerant structure design: what if we use the faulty power module in a new way, instead of getting rid of it using classic methods to disconnect it, i.e. systematically isolating the power device using fuses? The DLB power module was the perfect candidate to experience our philosophy. In the first place, a comparison of both technologies has been performed through post-fault-behaviour characterisation using controlled energy failure tests. Post-fault resistances, critical energies and overcurrent capability have been measured on a dedicated test-bench, along with defect localization and analysis through micro-section thanks to the CNES-THALES Lab in Toulouse, using non-intrusive Lock-In Thermography (LIT) method. Failed DLB power-modules have showed post-fault resistances 10 times lower than wire-bonded power-modules with the same die size and the same destruction energy. The clip also reduces temperature gradient around the defect location and thus, releases the resin’s thermal constraints compared to WB technology. The very thick solder joint between clip and chip ensures a better metallic reforming and therefore a less resistive post-fault resistance. In the second place, faulty power modules under low and high destruction energy, both WB and DLB, have been tested during 5 weeks for durability and robustness. Results clearly show DLB supremacy. Then, a long campaign of thermal characterization of both designs (Rth/Zth) has been carried out thanks to another dedicated bench. We have proposed a new heating technique setting the die in its linear mode, which avoids using a high current power supply. We have modelled both designs using COMSOL Multiphysics in order for them to be simulated and compared in terms of thermal resistance and impedance, electrical resistance and inductance. The DLB thermal diffuser effect has been analysed. Thermal resistances are very similar (~0,13°C/W) meanwhile, surprisingly, WB is better than DLB in terms of thermal impedance with a maximum difference of 20% at 0.1s.

Neste trabalho são caracterizados corpos sub-vulcânicos expostos no nordeste da Cordilheira Ocidental da Colômbia através de estudos petrográficos, análises geoquímicas em rocha-total e datação U-Pb por SHRIMP IIe em zircão. A área de estudo, em um contexto regional, representa um orógeno de acreção no sistema andino, constituído por rochas ígneas de afinidade oceânica amalgamadas contra a margem ocidental de América do Sul e associadas ao desenvolvimento da Grande Província Ígnea do Caribe (Caribbean Large igneous Province-CLIP). As análises em geoquímica elementar evidenciam para os corpos subvulcânicos um ambiente de subducção intra-oceânico de arco primitivo a mais evoluído, com significante contribuição de sedimentos subductados e assinaturas similares às registradas para as rochas vulcânicas e plutônicas presentes na área (Formação Barroso e Tonalito de Santa Fe e Buriticá). Dados isotópicos de Sr e Nd (razão inicial 87Sr/86Sr entre 0.70334 a 0.70543 e ?Nd entre +6.25 a +7.42) indicam uma área fonte de cunha mantélica, típica de rochas de arco, com níveis subordinados de material crustal. Idades U-Pb por SHRIMP IIe em zircão mostram cinco populações de idades correspondentes a: idades Neoarquenas ao redor de 2700 Ma, Paleoproterozoicas em torno de 2100, 2000 e 1800 Ma, Neoproterozoicas ao redor de 600 Ma, Ordovicianas/Devonianas de 460 e 360 Ma, e Cretáceas ao redor de 90 Ma. A população de idades mais novas é interpretada como a idade de cristalização dos corpos estudados, indicando uma relação espacial e temporal com as rochas ígneas da área associadas ao desenvolvimento da CLIP. As quatro populações de idades mais antigas são atribuídas a xenocristais, interpretados como provenientes de sedimentos terrígenos que foram arrastados ao canal de subducção, refletindo processos de reciclagem crustal através do manto em zonas de subducção intra-oceânicas. Foram identificadas duas possíveis áreas fontes para os xenocristais de zircão: o Complexo Marañon e o Maciço de Arequipa, que constituem inliers expostos na margem ocidental de América do Sul e apresentam uma boa correlação entre os eventos magmáticos/metamórficos que registram e às idades dos xenocristais encontrados. A partir destas interpretações, foi restringido o local de formação dos corpos estudados em relação à paleomargem continental, considerando além dados paleomagnéticos e modelos paleogeográficos publicados na zona de estudo. O trabalho fornece um aporte significativo para a compreensão da evolução tectono-magmática da região durante o Cretáceo superior, associada à história geológica da CLIP. / In this study sub-volcanic bodies exposed in the northeast of the Western Cordillera of Colombia are characterized through petrographic descriptions, geochemical analysis in bulk rocks and geochronology U-Pb SHRIMP in zircon. The area, in a regional context, represents an accretion orogen in the Andean system, it is composed of igneous rocks of oceanic affinity accreted to the western margin of South America, and they are associated to the development of the Caribbean Large igneous Province (CLIP). Based on geochemical characterization, the sub-volcanic bodies display patterns of an intra-oceanic subduction environment of primitive to evolved island arc, with significant contribution of subducted sediments, and similar signatures to those reported for the volcanic and plutonic rocks present in the area. (Barroso Formation and Santa Fe and Buriticá Tonalite). The Sr and Nd isotopic data (initial ratios 87Sr/86Sr between 0.70334 a 0.70543 and ?Nd between +6.25 a +7.42) suggest a wed mantle source, typical of arc rocks, with subordinates levels of crustal material. SHRIMP U-Pb data show five age populations that include: Neoarchean age around 2700 Ma, Paleoproterozoic age about of 2100, 2000 and 1800 Ma, Neoproterozoic age of 600 Ma, Ordovician-Devonian age of 460 and 360 Ma, and Cretaceous age around 90 Ma. The younger age population is interpreted as the crystallization age of the studied bodies; it indicated a special and temporal relation with the igneous rocks of the area which are associated with the development of the CLIP. The other four age populations are attributed to xenocrystals, they are interpreted as coming from terrigenous sediments dragged for the subduction area, reflecting process of crustal recycling through the mantle in intra-oceanic subduction zones. We restricted two possible source areas for the xenocrystals zircon: The Marañon Complex and The Arequipa Massif. They are constitute inliers exposed of the western margin of South America and present a good correlation between the magmatic/metamorphic events and the ages of the finding in our xenocristals. Based on these interpretations, we constraint the local of formation of the studied bodies closer to the continental paleomargin, considering also paleomagnetic data and paleogeographic models published in the area. The present study provides a significant contribution for the understanding tectono-magmatic evolution of the region during of Cretaceous, associated with the geological history of the CLIP.

Image generation has been taking large strides and new models showing great potential have been created. One of the continued struggles with image generation is controlling what the output will be, with no real way of guiding the generation into creating what the user wants. This has now been improved with the creation of text-image similarity models, which can be used together with an image generation model to guide the generation. This thesis will examine this new method of using a text-image similarity model and see how well it can generate pixel art of humanoid characters. The thesis compares the popular model Diffusion with a simple image generation method that relies solely on the text-image similarity models guidance. The results show that combining a diffusion model with a text-image similarity model improves the results over only using the text-image similarity model in almost every regard. Using a text-image similarity model allows the user to guide the generation, although sometimes the model will misinterpret the request. / Bildgeneration har tagit stora steg och nya modeller har tagits fram som visar stor potential. En av de forsatta svårigheterna med bildgeneration är att kontrollera vad modellen genererar. De nya text-bild-likhet modellerna förenklar nu för användare att tillsammans med en bildgenerator modell använda text-bild-likhet modellen att styra bildgeneratorn. Den här uppsatsen kommer utforska den nya metoden och se hur väl den kan användas för att generera mänskliga pixel art karaktärer. I uppsatsen kommer den populära Diffusion modellen jämföras med en enkel ritmetod som styrs av text-bild likhet modeller. Resultatet visar att kombinationen av en Diffusion modell och text-bild likhets modell ökar prestandan på nästan alla sätt i jämförelse med att låta text-bild-likhets modellen styra bildgeneratorn helt och hållet. Det visar sig att text-bild likhet modellen kan användas för att styra generationen men ibland så missförstår modellen vad som önskas.

RNA-bindende Proteine spielen eine zentrale Rolle in der posttranskriptionellen Kontrolle von mRNAs, die zwischen Transkription und Abbau von mRNAs stattfindet. RNA-bindende Proteine beeinflussen Spleißen, Export, Stabilität, Lokalisierung und Translation von mRNAs. FUS, EWSR1 und TAF15 gehören zu der Familie der FET Proteine. Diese wirken an verschiedenen zellulären Prozessen wie Transkription, Spleißen und der Prozessierung von miRNAs mit. Translokationen und Mutationen der FET Proteine führen zu verschiedenen Krankheiten. FUS spielt eine Rolle bei den neurodegenerativen Krankheiten frontotemporale Lobärdegeneration (FTLD) und amyotrophe Lateralsklerose (ALS). In dieser Arbeit wurde die mithilfe von photoaktivierbaren Ribonukleotiden UV-Licht induzierte Quervernetzung und Immunpräzipitation (PAR-CLIP) Methode genutzt, um die RNA-Bindestellen von FUS, EWSR1 und TAF15, einer ALS-verursachenden FUS Mutante und einem anderen, mit ALS in Verbindung stehenden Protein, TARDBP, zu bestimmen. Die RNA-Bindestellen der FET-Proteine lagen größtenteils in Introns. Passend dazu konnte durch knockdown der FET Proteine eine Rolle von FUS und EWSR1 im Spleißen von mRNAs validiert werden. Dem Ubiquitin-Proteasom-System zugehörige RNAs waren unter den sowohl von FUS als auch TARDBP gebundenen mRNAs überrepräsentiert. Dies bestätigt die Annahme, dass Störungen in der Proteindegradation die ALS-Pathogenese beeinflussen. Zusätzlich konnte gezeigt werden, dass FUS und TAF15 bevorzugt UAC-reiche, einzelsträngige RNA-Sequenzen binden. Sequenzierung von mRNAs nach Depletion von FUS, EWSR1 und TAF15 in HEK293-Zellen zeigte einen stabilisierenden Effekt der FET-Proteine auf gebundene mRNAs. Desweiteren scheinen die FET Proteine durch Interaktion mit Promotor-assoziierten, nicht-kodierenden RNAs die Transkription zu beeinflussen. / Post-transcriptional regulation of gene expression takes place at multiple levels between transcription and decay of the mRNA. RNA-binding proteins play a key role in orchestrating splicing, export, stability, localization and translation of mRNAs. FUS, EWSR1 and TAF15 constitute the FET protein family which participates in multiple levels of cellular function. FET proteins have been implicated to function in various cellular processes including transcription, pre-mRNA splicing and miRNA processing. Translocations and mutations in FET proteins lead to diverse pathologies. FUS is involved in neurodegenerative diseases like frontotemporal lobar degeneration (FTLD) and amyotrophic lateral sclerosis (ALS). In this study, Photoactivatable-Ribonucleoside-Enhanced Crosslinking and Immunoprecipitation (PAR-CLIP) was used to determine RNA-targets and binding sites of FUS, EWSR1 and TAF15, an ALS-causing FUS mutant and another ALS-related protein, TARDBP. The identified binding sites of FET proteins were mainly intronic, supporting the involvement of FUS and EWSR1 in splicing, which was validated by FET protein knockdown. Comparison of FUS and TARDBP RNA targets revealed that ubiquitin-proteasome related gene categories were overrepresented, further illustrating that aberrations in protein degradation are implicated in the pathogenesis of ALS. In addition, it was shown that FUS and TAF15 proteins preferentially bind UAC rich, single-stranded RNA sequences. mRNA sequencing after FUS, EWSR1 and TAF15 depletion in HEK293 cells revealed a stabilizing effect on their targets. Interestingly, FET proteins also seem to influence transcription by interaction with promoter-associated noncoding RNAs. In summary, we identified the RNA-targets and binding sites of all human FET proteins in comparison with an ALS-causing FUS mutant and TARDBP. Functional studies revealed an involvement of FET proteins in mRNA stabilization, splicing and transcriptional regulation.

Le syndrome de l’X fragile représente la première cause de déficience intellectuelle héréditaire. Ce syndrome résulte de l’absence de la protéine FMRP. FMRP est proposée réguler, sous contrôles des mGluR-I et d’autres récepteurs, l’expression de protéines importantes pour la plasticité synaptique en se fixant spécifiquement sur leur ARNm et en modulant leur traduction. Des milliers d’ARNm cibles ont déjà été proposées dans la littérature, mais très peu ont pu être validées. Par approche de pontage covalent aux UV et immunoprecipitation (CLIP) couplé à une analyse microarray, nous avons identifié un ARNm comme cible unique de FMRP dans les neurones corticaux. Cet ARNm code pour une kinase contrôlant le niveau de deux seconds messagers lipidiques importants pour le remodelage des épines dendritiques. De plus, nous avons montré que l’activation mGluR-I dépendante de la kinase est absente dans les neurones Fmr1 KO, avec pour conséquence une altération de plusieurs espèces lipidiques du neurone. Ces défauts peuvent expliquer les altérations morphologiques et fonctionnelles des épines dendritiques, cause principale proposée du syndrome de l’X fragile. / Fragile X syndrome is the leading cause of inherited intellectual disability and is due to the absence of the RNA binding protein FMRP (Fragile X Mental Retardation Protein). FMRP is proposed to bind and regulate synaptic expression of mRNA targets upon mGluR-I activation. Thousands of mRNA targets have already been proposed in the literature, but only a few have been validated leaving unsolved the question of the genes mostly affected by the absence of FMRP in the brain of fragile Xpatients. The main project of the thesis was to identify the mRNAs associated with FMRP in cortical neurons by performing cross-linking immunoprecipitation approach (CLIP). We found that FMRP principally targets one unique mRNA which encodes an important synaptic kinase. This enzyme controls the level of two second lipid messengers important for remodeling of dendritic spines. Consequently, the mGluR-I-dependant activation of the enzyme is lost in absence of FMRP, leading to several lipid species alterations in the neuron. These defects may explain the morphological and functional alterations of dendritic spines, the hallmark of fragile X syndrome.

Die chloroplastidäre und mitochondriale Genexpression ist abhängig von einer großen Anzahl an RNA-Bindeproteinen (RBPs). Eine besonders abundante Familie sind die chloroplastidären Ribonukleoproteine (cpRNPs). Während Ziel-RNAs mehrerer cpRNPs und die Phänotypen entsprechender Mutanten beschrieben wurden, bleibt ihre molekulare Funktion weitgehend ungeklärt. In dieser Arbeit wurden Studien der cp29a Mutante durch genomweite Analysen erweitert. Diese legen nahe, dass die eigentliche Rolle von CP29A in phänotypisch erkennbarem Mutanten-Gewebe durch sekundäre Defekte maskiert wird. Um primäre Defekte zu identifizieren, wurden in vivo Bindestellen von CP29A mit einer neuen Chloroplasten-adaptierten Methode, die UV-Licht zur Quervernetzungen nutzt, bestimmt. Transkripte, die für Untereinheiten des Photosystem II und des Cytochrom-b6f-Komplexes kodieren, waren unter den Zielen von CP29A überrepräsentiert. Weiterhin wurden mehrere Bindestellen in Nachbarschaft zu Bindestellen von PPR-Proteinen identifiziert. Mit einer alternativen Methode, die chemische Quervernetzung nutzt, wurden Ziel-RNAs eines weiteren cpRNP, CP31A, identifiziert. Transkripte, die für Untereinheiten des NADH-Dehydrogenase Komplexes kodieren, waren überrepräsentiert. Diese Daten führten zu einer neuen Hypothese, die die Funktion von cpRNPs im Zusammenspiel mit PPR-Proteinen in der Prozessierung funktionell verwandter RNAs postuliert. Ein weiterer für die Genexpression relevanter Mechanismus ist die Bildung membranloser Kompartimente durch flüssig-flüssig Phasentrennung. Es wurde eine in silico Analyse durchgeführt, um organelläre Proteine mit Domänen, die auf flüssig-flüssig Phasentrennung hindeuten, zu identifizieren. Funktionen mit Bezug zu Genexpression, insbesondere RNA-Edierung, waren bei diesen Proteinen mit Prionen-ähnlichen Domänen (PLDs) überrepräsentiert. Zwei Kandidaten wurden auf ihre Neigung zur flüssig-flüssig Phasentrennung durch in vitro Experimente und in vivo Mikroskopie untersucht. / Gene expression in chloroplasts and mitochondria relies on a large number of RNA-binding proteins (RBPs), which are involved in the processing of polycistronic precursor transcripts. A particular abundant family are the chloroplast ribonucleoproteins (cpRNPs). While target RNAs and mutant phenotypes of several cpRNPs were described, insights on their molecular function remained sparse. In this thesis, analyses of cp29a mutants were extended by genome-wide transcriptome data, which suggest that in phenotypically noticeable mutant tissue the actual role of CP29A might be masked by secondary effects. To identify primary defects, in vivo binding sites of CP29A on its target transcripts were determined using a novel chloroplast-adapted approach using crosslinking by UV-light. Identified targets of CP29A are functionally enriched in mRNAs encoding subunits of the photosystem II and the cytochrome b6f complex. Moreover, several binding sites were identified in close proximity to characterized binding sites of PPR proteins. Using an alternative approach, employing chemical crosslinking, targets of another cpRNP, CP31A, were identified. Targets are enriched in genes encoding subunits of the NADH-like dehydrogenase complex. In combination, these data led to a novel hypothesis on the molecular function of cpRNPs working together with PPR proteins in the processing of functionally related RNAs. Another increasingly recognized mechanism in gene expression is the formation of membraneless organelles by liquid-liquid phase separation. An in silico screen for organellar proteins containing domains indicative of phase separation was performed. The identified set of proteins with prion-like domains (PLDs) is enriched in functions related to gene expression, particularly RNA-editing. Two selected candidate proteins were characterized for their propensity to undergo phase separation by in vitro phase separation assays and in vivo microscopy.

Female adults of many mosquito species possess distinct physiological features adapting to blood feeding for successful reproduction. The disease pathogens that are transmitted by mosquitoes have evolved to take advantages of the indispensable blood feedings to complete their transmission cycles and to survive attacks from the mosquito's innate immune system. Normal egg development and mosquito immunity are tightly controlled by tissue- and stage-specific gene expression and coordinated by many signal molecules in the mosquito. Understanding gene regulation affecting mosquito reproduction and malaria parasites infection is of paramount importance for developing novel malaria control strategies. A growing body of evidence indicates that microRNAs (miRNAs) are involved in egg maturation and immune reactions against invading pathogens in mosquitoes. However, the molecular mechanisms by which specific miRNAs selectively modulate reproduction and the survival of pathogens are largely unknown. The miRNA-induced gene-silencing pathway in mosquitoes was mostly extrapolated from the studies of flies. To explore the dynamics of miRNAs in reproduction, I used small RNAs sequencing to monitor miRNAs expression and their association with Argonaute 1 (Ago1) and Argonaute 2 (Ago2) in the malaria mosquito Anopheles gambiae (An. gambiae) during the 72-h period immediately after blood feeding. I found the abundance and Ago loading of most of the mature miRNAs were relatively stable after blood ingestion. However, miRNAs of the miR-309/286/2944 cluster were considerably upregulated after blood feeding. I confirmed that miR-309 is essential for normal egg development by depletion of endogenous miR-309 with a specific antagomir. In addition, my results showed that the Ago association of some miRNAs was not proportional to their cellular abundance implying additional regulation at miRNA integration. To investigate the functional roles of miRNAs and define context-dependent miRNA-mRNA interactions during the reproductive process, I have applied an innovative experimental approach to study miRNA-mRNA interactome. CLEAR (covalent ligation of endogenous Argonaute-bound RNAs)-CLIP can generate miRNA-mRNA chimeras from UV-irradiation stabilized Ago-miRNA-mRNA complex. My results have defined tens of thousands of miRNA-mRNA interactions in mosquitoes, including novel targets for mosquito-specific miRNAs. Verification of the predicted interactions using mRNA-seq, ribosome-profiling, and luciferase reporter assay revealed a reliable miRNA-mRNA interaction network. Based on the detected interactions, I refined the paring rules for mosquito miRNAs and illustrated the dynamic pairing between different regions of miRNAs with their targets in vivo. The miRNA-mRNA interactions were compared using this approach at multiple time points before and after blood feeding. Importantly, this study showed that the interactions were dynamic and enriched in genes that are involved in metabolisms, supporting the proposed functions of miRNAs in coordinating the gene regulation in mosquito reproduction. Plasmodium falciparum (P. falciparum) is a major human malaria parasite. To understand the functions of miRNAs in the mosquito resistance to Plasmodium infection, we analyzed the miRNA-mRNA interactions after female mosquitoes taking a P. falciparum-infected blood meal or an uninfected blood meal. Comparison of the interactions revealed enhanced miRNA-mRNA interactions after P. falciparum infection involving a group of immunity-related genes. In summary, this study has provided a systematic view and significantly advanced our understanding of the miRNA functions in mosquito reproduction and P. falciparum infection. / PHD / Female mosquito is able to transmit lots of disease to the human when it bites for blood. The blood meal provides necessary nutrient for mosquito reproduction and spread the pathogens such as malaria and Zika at the same time. Thus understanding the molecular mechanism behind this process would be greatly helpful to develop novel vector control strategy. Here, we found a distinct class of RNAs contributing to the regulation of mosquito blood meal and parasite infection. We used a novel biochemical method to decoding the special role of these kinds of RNAs in these processes. We found them regulating mosquito metabolism and immunity. This study significantly deepened our knowledge about the process of mosquito reproduction and transmitting diseases.

Enabling machines to solve computer vision tasks with natural language components can greatly improve human interaction with computers. In this thesis, we address vision and language tasks with deep learning methods that explicitly localize relevant visual evidence. Spatial evidence localization in images enhances the interpretability of the model, while temporal localization in video is necessary to remove irrelevant content. We apply our methods to various vision and language tasks, including visual question answering, temporal activity detection, dense video captioning and cross-modal retrieval. First, we tackle the problem of image question answering, which requires the model to predict answers to questions posed about images. We design a memory network with a question-guided spatial attention mechanism which assigns higher weights to regions that are more relevant to the question. The visual evidence used to derive the answer can be shown by visualizing the attention weights in images. We then address the problem of localizing temporal evidence in videos. For most language/vision tasks, only part of the video is relevant to the linguistic component, so we need to detect these relevant events in videos. We propose an end-to-end model for temporal activity detection, which can detect arbitrary length activities by coordinate regression with respect to anchors and contains a proposal stage to filter out background segments, saving computation time. We further extend activity category detection to event captioning, which can express richer semantic meaning compared to a class label. This derives the problem of dense video captioning, which involves two sub-problems: localizing distinct events in long video and generating captions for the localized events. We propose an end-to-end hierarchical captioning model with vision and language context modeling in which the captioning training affects the activity localization. Lastly, the task of text-to-clip video retrieval requires one to localize the specified query instead of detecting and captioning all events. We propose a model based on the early fusion of words and visual features, outperforming standard approaches which embed the whole sentence before performing late feature fusion. Furthermore, we use queries to regulate the proposal network to generate query related proposals. In conclusion, our proposed visual localization mechanism applies across a variety of vision and language tasks and achieves state-of-the-art results. Together with the inference module, our work can contribute to solving other tasks such as video question answering in future research.

RNA helicases are found in all kingdoms of life, functioning in all aspects of RNA biology mainly through modulating structures of RNA and ribonucleoprotein (RNP) complex. RNA structures have fundamental impacts on steps in gene expression, including transcription, pre-mRNA processing, and translation. However, the precise roles and regulatory mechanisms of RNA structures in co- and post-transcriptional processes remain elusive. By probing genome-wide RNA structures in vivo, a recent study suggested that ATP-dependent factors, such as RNA helicases, maintain the actively unfolded state of RNAs. Among all RNA helicases, DEAD-box proteins form the largest family in eukaryotes, and have been shown to remodel RNA/RNP structures both in vitro and in vivo. Nevertheless, for the majority of these enzymes, it is largely unclear what RNAs are targeted and where they modulate RNA/RNP structures to regulate co-transcriptional processes. To fill the gap, my research focused on identification of the RNAs and structures targeted by the DEAD-box protein Dbp2 in S. cerevisiae to uncover the cellular processes that Dbp2 is involved in.<br><div><div>My studies revealed a role of Dbp2 in transcriptional termination. Dbp2 binds to ~34% of yeast mRNAs and all snoRNAs, and loss of DBP2 leads to a termination defect as evidenced by RNA polymerase II (RNAPII) accumulation at 3’ ends of these genes. In addition, the binding pattern of Dbp2 in mRNAs is highly similar to Nrd1 and Nab3 in the Nrd1-Nab3-Sen1 (NNS) termination complex, and deletion of DBP2 leads to reduced recruitment of Nrd1 to its target genomic loci. In Dbp2 and NNS targeted 3’ UTRs, RNA structural changes resulted from DBP2 deletion also overlap polyadenylation elements and correlate with inefficient termination, and loss of stable structure in the 3’ UTR bypasses the requirement for Dbp2. These findings lead to a model that Dbp2 promotes efficient termination of transcription through RNA structure remodeling.</div><div>Interestingly, my research also revealed the requirement of DBP2 for efficient splicing, as loss of DBP2 leads to accumulation of unspliced pre-mRNAs. Moreover, this function is dependent on the helicase activity of Dbp2. Further studies are needed to characterize the molecular mechanism of how Dbp2 facilitates splicing in cells. Overall, my research demonstrated that DEAD-box RNA helicases remodel mRNA structure in vivo and that structural alteration can be essential for proper gene expression.</div></div>

Made available in DSpace on 2016-03-15T19:46:54Z (GMT). No. of bitstreams: 1 Bianca Rodrigues dos Santos.pdf: 912851 bytes, checksum: 3a73fc4f8e721621c7704a7a734b8387 (MD5) Previous issue date: 2009-08-20 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / With the invention and popularization of the television, the radio has partly lost its space. However, the song, defined in this work as the junction of lyrics and music, has continued to be part of people s lives, as it is nowadays. This way, every country has its national anthem and every urban tribe appreciates one different kind of song, which contributes fundamentally for the identity s construction of this specific group. Therefore, there has been the need of a language that joined the song, which is sonorous, with the image, in other words, an audiovisual language. In the 1980 s appears a new language: the video clip. Despite being limited by the length of the song, this new language presents a great creative capacity and its directors have, at hand, the most recent technological resources from the current digital era. The video clip is part of the cultural industry and it can be considered a modern language. Thereby, this essay s objective is to analyze, from a corpus made of three Brazilian video clips from the 1990 s, how the video clip (re)constructs the song. The analyses will be done under different aspects: in Segue o seco , interpreted by Marisa Monte, there is the Brazilian northeastern theme; in Amor I love you , from the same singer, there is the dialogue with the plot of O Primo Basílio; and in Seus Passos , interpreted by the band Skank, the dialogue with the surrealism. / Com o advento e popularização da televisão, o rádio perdeu parte de seu espaço. Entretanto a canção, definida neste trabalho como a soma de letra e música, tem continuado a fazer parte da vida de todas as pessoas. Assim, todo país tem seu hino nacional e toda tribo urbana aprecia um tipo de canção diferente, o que contribui, de maneira fundamental, para a construção da identidade desse determinado grupo. Ao longo o tempo, houve a necessidade de uma linguagem que unisse a canção, que é sonora, com a imagem, ou seja, uma linguagem audiovisual. Surge, então, a partir da década de 80 do século XX, o videoclipe. Apesar de delimitada pelo tempo da canção, essa nova linguagem apresenta uma grande capacidade criativa e seus diretores têm em mãos os mais recentes recursos provenientes da era digital na qual vivemos. O videoclipe faz parte da indústria cultural e pode ser considerado como uma linguagem moderna. Este trabalho tem como objetivo analisar, a partir de um corpus composto por três videoclipes nacionais da década de 90, como o videoclipe (re)constrói a canção. As análises são realizadas sob diferentes aspectos: em Segue o seco , interpretada por Marisa Monte, há a temática da seca nordestina; em Amor I love you , da mesma intérprete, o diálogo com a obra O Primo Basílio; e em Seus Passos , interpretada pela banda Skank, o diálogo com o surrealismo.