Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2017-03-28T13:13:15Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Cabral Filho, PE Tese - PPGCB.pdf: 40667296 bytes, checksum: 0fe805957d5ab95d53bed853da61deb5 (MD5) / Made available in DSpace on 2017-03-28T13:13:15Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Cabral Filho, PE Tese - PPGCB.pdf: 40667296 bytes, checksum: 0fe805957d5ab95d53bed853da61deb5 (MD5) Previous issue date: 2016-07-15 / CAPES / Os pontos quânticos ou quantum dots (QDs) são nanocristais fluorescentes de semicondutores com propriedades ópticas únicas, tendo como principais vantagens: (1) alta resistência à fotodegradação, possibilitando o acompanhamento de eventos biológicos em tempo real e, (2) superfície ativa, permitindo a conjugação a biomoléculas que vão propiciar especificidade às marcações, além de possibilitar também sua ligação a outras nanopartículas. Com isso, é possível quantificar uma variedade de biomoléculas em células e tecidos e desenvolver nanossondas bimodais (magnético-fluorescentes) baseadas em QDs. O desenvolvimento de nanopartículas bimodais pode aliar as vantagens das técnicas baseadas em fluorescência com as de imagem por ressonância magnética (IRM). Entretanto, a obtenção de sondas bimodais é ainda um desafio, pois durante a conjugação devem ser mantidas as propriedades fluorescentes e magnéticas das nanopartículas, e com isso ainda há poucos trabalhos que façam aplicações em sistemas biológicos. O objetivo desta tese se caracteriza pelo desenvolvimento de sondas com propriedades multimodais baseadas em QDs de Telureto de Cádmio (CdTe) associadas a nanopartículas magnéticas de óxido de ferro como marcadores sítio-específicos em células cancerígenas. Inicialmente os QDs foram conjugados covalentemente à transferrina (Tf) [QDs-Tf] para a quantificação específica de seus receptores (TfRs) em células HeLa e em duas linhagens de glioblastoma (U87 e DBTRG). Através de ensaios de saturação do TfR, foi possível inferir sobre a taxa de renovação deste receptor nessas células. Os resultados mostraram que as células HeLa e as DBTRG possuem uma maior quantidade do TfR quando comparadas às U87. As DBTRG apresentaram maior taxa de renovação do TfR, quando comparadas aos outros dois tipos, demonstrando que os conjugados QDs-Tf são potenciais ferramentas para o estudo da biologia celular do câncer. Posteriormente, nanossondas bimodais (QDsMNPs), baseadas em QDs associados a nanopartículas magnéticas de óxido de ferro, foram obtidas por conjugação covalente. De acordo com as caracterizações, QDs-MNPs mantiveram suas propriedades ópticas e magnéticas e apresentaram-se como potenciais sondas inespecíficas para fluorescência e para aquisição de imagens por RM ponderadas em T2 (tempo de relaxação nuclear transversal). A conjugação prévia dos QDs a Tf, além de fornecer informações sobre a biologia do câncer, auxiliou também na padronização da marcação específica do TfR em células cancerígenas e no estabelecimento de protocolos de conjugação das sondas bimodais a Tf. Por fim, as QDs-MNPs foram conjugadas covalentemente a Tf e essa nova sonda multimodal [(QDs-MNPs)-Tf] reconheceu especificamente os TfR em células HeLa. As caracterizações indicaram que o sistema multimodal não apresentou alteração significativa nas propriedades ópticas e exibiu uma maior relaxividade transversal (r2), se mostrando igualmente potencial sonda para análise por fluorescência e IRM ponderada em T2. Neste trabalho foram obtidas nanossondas promissoras para serem aplicadas na compreensão da biologia celular do câncer, além de auxiliar em métodos diagnósticos e terapêuticos para essa doença. / Quantum dots (QDs) are fluorescent semiconductor nanocrystals with unique optical properties, which have as major advantages: (1) the high resistance to photobleaching, making possible to monitor biological events in real-time and, (2) active surface, allowing the conjugation not only with biomolecules for specific labeling, but also to other nanoparticles. Thus, it would be possible to quantify a variety of biomolecules in cells and tissues, as well as to develop bimodal nanoprobes (fluorescent-magnetic) [BNPs] based on QDs. The development of BNPs can help to combine the advantages of the fluorescence with the resonance magnetic imaging techniques. However, the preparation of bimodal probes can still be considered a challenge, since the fluorescent and magnetic nanoparticles’ properties need to be preserved after conjugation. Therefore, there are still few works applying BNPs in biological studies. The aim of this thesis was to develop nanoprobes, with multimodal properties, based on cadmium telluride (CdTe) QDs conjugated with iron oxide magnetic nanoparticles (MNPs), for site-specific labeling in cancer cells. For this, initially, QDs were covalently coupling to transferrin (Tf) [QDs-Tf] and used to quantify the transferrin receptor (TfRs) in HeLa cells as well as in two glioblastoma lines (U87 and DBTRG). Furthermore, by a TfR saturation assay, it was possible to study the recycling rate of this receptor in cells studied. The results showed that HeLa and DBTRG cells present a higher amount of TfRs when compared to U87. DBTGR showed a higher TfR recycling rate, when compared to the other two lineages, demonstrating that QDs-Tf conjugates are potential tools to study the cancer cell biology. BNPs, based on the conjugation of QDs with MNPs (QDs-MNPs), were obtained by covalent coupling. According to characterizations, the BNPs remained with their optical and magnetic properties preserved and showed to be potential unspecific probes for fluorescence analysis and for T2-weighted magnetic resonance imaging (MRI) acquisition. The conjugation of QDs to Tf, performed previously, was a valuable step not only to provide us information about the biology of cancer cells, but also for the standardization of TfR specific labeling and the establishment of protocol to conjugate the BNPs with Tf. Therefore, QDs-MNPs were also covalently coupling to Tf and this new multimodal nanotool [(QDs-MNPs)Tf] was also able to recognize specifically TfRs in HeLa cells. The multimodal nanosystems presented their fluorescent properties practically unchanged and also exhibited a higher transversal relaxivity (r2), when compared to bare BNPs, showing likewise potential to be used for fluorescence and T2-weighted MRI analyses. In this work, it was developed promising nanoprobes, able to be applied for the cancer cell biology comprehension, and with potential for helping in the improvement of diagnostic and therapeutic methods for this disease.

The rapid expansion of artificial intelligence (AI) in modern society, exemplified by systems like ChatGPT and Stable Diffusion, has given rise to significant ethical considerations. These systems, increasingly prevalent in diverse sectors such as mental health treatment, as in Koko, and art creation, necessitate a careful examination of their alignment with human values. This thesis addresses the pressing need for ethical evaluation of multimodal AI systems - those capable of processing and responding to both text and image inputs. Our research is twofold: initially, we focus on developing a multimodal ethical database through interactive human feedback. Participants assess various examples, determining their ethical appropriateness. This process culminates in a dataset that serves as a foundation for the subsequent phase - designing and testing algorithms capable of autonomously evaluating the morality of AI responses. We explore the effectiveness of two models in this context: a RoBERTa-large classifier and a multilayer perceptron classifier. Furthermore, this thesis highlights significant limitations in the existing multimodal AI systems studied. We propose alternative models, offering a comparative analysis mainly in terms of performance. This comprehensive study not only contributes to the field of AI alignment but also proposes methodologies for enhancing the moral framework within which these influential technologies operate. / L'expansion rapide de l'intelligence artificielle (IA) dans la société moderne, illustrée par des systèmes tels que ChatGPT et Stable Diffusion, a suscité d'importantes considérations éthiques. Ces systèmes, de plus en plus présents dans divers secteurs tels que le traitement de la santé mentale, avec Koko, et la création artistique, nécessitent un examen attentif de leur alignement avec les valeurs humaines. Ce mémoire aborde le besoin pressant d'une évaluation éthique des systèmes d'IA multimodaux - capables de traiter et de répondre à la fois aux entrées textuelles et visuelles. Notre recherche est double : initialement, nous nous concentrons sur le développement d'une base de données éthiques multimodales par le biais de retours interactifs d'utilisateurs. Les participants évaluent divers exemples pour déterminer leur éthique. Ce processus aboutit à un ensemble de données qui sert de fondement à la phase suivante - la conception et le test d'algorithmes capables d'évaluer de manière autonome la moralité des réponses de l'IA. Nous explorons l'efficacité de deux modèles dans ce contexte : un classificateur RoBERTa-large et un perceptron multicouche. De plus, ce mémoire met en évidence des limitations significatives dans les systèmes d'IA multimodaux existants étudiés. Nous proposons des modèles alternatifs, offrant une analyse comparative en termes de performance. Cette étude complète contribue non seulement au domaine de l'alignement de l'IA, mais propose également des méthodologies pour améliorer le cadre moral dans lequel ces technologies influentes opèrent.