As Next Generation (NextG) networks become more complex, the need to develop a robust, reliable network traffic prediction framework for intelligent network management increases. This study compares the performance of machine learning models in network traffic prediction using a custom Service-Level Agreement (SLA) - based loss function to ensure SLA violation constraints while minimizing overprovisioning. The proposed SLA-based parametric custom loss functions are used to maintain the SLA violation rate percentages the network operators require. Our approach is multivariate, spatiotemporal, and SLA-driven, incorporating 20 Radio Access Network (RAN) features, custom peak traffic time features, and custom mobility-based clustering to leverage spatiotemporal relationships. In this study, five machine learning models are considered: one recurrent neural network (LSTM) model, two encoder-decoder architectures (Transformer and Autoformer), and two gradient-boosted tree models (XGBoost and LightGBM). The prediction performance of the models is evaluated based on different metrics such as SLA violation rate constraints, overprovisioning, and the custom SLA-based loss function parameter. According to our evaluations, Transformer models with custom peak time features achieve the minimum overprovisioning volume at 3% SLA violation constraint. Gradient-boosted tree models have lower overprovisioning volumes at higher SLA violation rates. / Master of Science / As the Next Generation (NextG) networks become more complex, the need to develop a robust, reliable network traffic prediction framework for intelligent network management increases. This study compares the performance of machine learning models in network traffic prediction using a custom loss function to ensure SLA violation constraints. The proposed SLA-based custom loss functions are used to maintain the SLA violation rate percentages required by the network operators while minimizing overprovisioning. Our approach is multivariate, spatiotemporal, and SLA-driven, incorporating 20 Radio Access Network (RAN) features, custom peak traffic time features, and mobility-based clustering to leverage spatiotemporal relationships. We use five machine learning and deep learning models for our comparative study: one recurrent neural network (RNN) model, two encoder-decoder architectures, and two gradient-boosted tree models. The prediction performance of the models was evaluated based on different metrics such as SLA violation rate constraints, overprovisioning, and the custom SLA-based loss function parameter.

La capacité des lectines à reconnaître spécifiquement des glycoconjugués à la surface de cellules en font des outils de diagnostic biomédical pour les pathologies associées à des changements de glycosylation (inflammation, du cancer ...). De par leur interaction avec les glycosphingolipides, ces protéines peuvent aussi être utilisées pour étudier le trafic membranaire. Toutefois, un nombre réduit de lectines sont actuellement disponibles, limitant leur utilisation dans les biotechnologies et la recherche. Le but de ma thèse est d'une part de concevoir des néo-lectines de valence et topologie contrôlées pour comprendre l'effet de la multivalence sur le mécanisme d'endocytose, et d'autre part de concevoir des lectines de spécificité modulable afin de les utiliser dans la reconnaissance spécifique des cellules tumorales.RSL est une lectine à fucose de la bactérie Ralstonia solanacearum qui a une structure en β-propeller formée par l'association de trois monomères présentant deux sites de liaison très similaires. Cette protéine trimérique et hexavalente a été choisie comme structure de base pour la conception de néolectines. Des RSLs trivalentes ont été produites par mutation d'un acide aminé essentiel pour la stabilisation du fucose. Leur caractérisation a démontré qu'ils avaient perdu la capacité d'invaginer la membrane plasmique. Une protéine de même structure que RSL mais monomérique a été ingénierée, puis une librairie de plus de 13 mutants de valence présentant différentes topologies a été créée. L'analyse de tous les mutants a permis de démontrer que la formation de tubules dans les membranes dépend plus de la distance entre les sites que du nombre de sites.Nous avons ensuite mis au point un protocole de bio-informatique afin de prédire l'orientation et la conformation d'oligosaccharides fucosylés dans les sites de fixation de plusieurs lectines à fucose. Les affinités relatives ont pu être calculées avec une bonne corrélation avec les valeurs expérimentales. La modélisation et la structure cristallographique des complexes entre RSL et les oligosaccharides Lewis X et Sialyl Lewis X indiquent un changement conformationnel du glycanne très inhabituel lors de l'interaction, donnant ainsi des pistes pour la conception de mutants de plus haute spécificité. / The ability of lectins to specifically recognize glycoconjugates on cell surface makes them excellent biomedical diagnostics tools for diseases associated with glycosylation changes (e.g inflammation, cancer, etc.). Furthermore, because of their interaction with glycosphingolipids, lectins may also be used to study membrane trafficking. However, only small number of lectins are currently available, limiting their use in biotechnology and research. The aim of my thesis was first to develop neolectins with controlled valency and topology to understand the effect of multivalency on the endocytosis mechanism, and second to design lectins with tuned specificity for the recognition of tumor cells.RSL is a fucose binding lectin from the bacterium Ralstonia solanacearum which has a β-propeller structure that is formed by the association of three monomers each having two very similar binding sites. This trimeric and hexavalent protein was chosen as the scaffold structure for the design of neolectins. Trivalent RSLs were created by mutating an amino acid with essential role in fucose binding. Characterization showed that these mutants lost the ability to invaginate the plasma membrane. In addition, monomeric RSL was engineered and a library of more than 13 mutants, with different topologies and valencies, was created. Analysis of these mutants showed that the formation of tubules in the membrane depends mostly on the distance between the sites rather than on the number of sites.Then we developed a bioinformatic protocol to predict the orientation and conformation of fucosylated oligosaccharides in the binding sites of several fucose binding lectins. The relative affinities could be calculated with a good correlation to experimental values. Both the model and the crystal structures of RSL complexed with sialyl Lewis X and Lewis X oligosaccharides indicate a very unusual conformational change of the glycan during the interaction. These studies pave the way for the design of mutants with higher specificity.