Les biotechnologies sont arrivées au point ou la quantité d'information disponible permet de penser les objets biologiques comme des systèmes complexes. Dans ce contexte, les phénomènes qui émergent de ces systèmes sont intimement liés aux spécificités de leur organisation. Cela pose des problèmes computationnels et statistiques qui sont précisément l'objet d'étude de la communauté liée à l'apprentissage statistique. Cette thèse traite d'applications de méthodes d'apprentissage pour l'étude de phénomène biologique dans une perspective de système complexe. Ces méthodes sont appliquées dans le cadre de l'analyse d'interactions protéine-ligand et d'effets secondaires, du phenotypage de populations de cellules et du plan d'expérience pour des systèmes dynamiques non linéaires partiellement observés.D'importantes quantités de données sont désormais disponibles concernant les molécules mises sur le marché, tels que les profils d'interactions protéiques et d'effets secondaires. Cela pose le problème d'intégrer ces données et de trouver une forme de structure sous tendant ces observations à grandes échelles. Nous appliquons des méthodes récentes d'apprentissage non supervisé à l'analyse d'importants jeux de données sur des médicaments. Des exemples illustrent la pertinence de l'information extraite qui est ensuite validée dans un contexte de prédiction.Les variations de réponses à un traitement entre différents individus posent le problème de définir l'effet d'un stimulus à l'échelle d'une population d'individus. Par exemple, dans le contexte de la microscopie à haut débit, une population de cellules est exposée à différents stimuli. Les variations d'une cellule à l'autre rendent la comparaison de différents traitement non triviale. Un modèle génératif est proposé pour attaquer ce problème et ses propriétés sont étudiées sur la base de données expérimentales.A l'échelle moléculaire, des comportements complexes émergent de cascades d'interactions non linéaires entre différentes espèces moléculaires. Ces non linéarités engendrent des problèmes d'identifiabilité du système. Elles peuvent cependant être contournées par des plans expérimentaux spécifiques, un des champs de recherche de la biologie des systèmes. Une stratégie Bayésienne itérative de plan expérimental est proposée est des résultats numériques basés sur des simulations in silico d'un réseau biologique sont présentées. / Biotechnologies came to an era where the amount of information one has access to allows to think about biological objects as complex systems. In this context, the phenomena emerging from those systems are tightly linked to their organizational properties. This raises computational and statistical challenges which are precisely the focus of study of the machine learning community. This thesis is about applications of machine learning methods to study biological phenomena from a complex systems viewpoint. We apply machine learning methods in the context of protein-ligand interaction and side effect analysis, cell population phenotyping and experimental design for partially observed non linear dynamical systems.Large amount of data is available about marketed molecules, such as protein target interaction profiles and side effect profiles. This raises the issue of making sense of this data and finding structure and patterns that underlie these observations at a large scale. We apply recent unsupervised learning methods to the analysis of large datasets of marketed drugs. Examples show the relevance of extracted information which is further validated in a prediction context.The variability of the response to a treatment between different individuals poses the challenge of defining the effect of this stimulus at the level of a population of individuals. For example in the context High Content Screening, a population of cells is exposed to different stimuli. Between cell variability within a population renders the comparison of different treatments difficult. A generative model is proposed to overcome this issue and properties of the model are investigated based on experimental data.At the molecular scale, complex behaviour emerge from cascades of non linear interaction between molecular species. These non linearities leads to system identifiability issues. These can be overcome by specific experimental plan, one of the field of research in systems biology. A Bayesian iterative experimental design strategy is proposed and numerical results based on in silico biological network simulations are presented.

Cancer cells have been known for some time to have very different metabolismas compared to that of normal non proliferating cells. As metabolism is involvedin almost every aspect of cell function, there has been a recent resurgence ofinterest in inhibiting cancer metabolism as a therapeutic strategy. Inhibitors thatspecifically target altered metabolic components in cancer cells are being developedas antiproliferative agents. However, many such inhibitors have not progressedinto the clinic due to limited efficacy either in vitro or in vivo. In this study weexplore the hypothesis that this is often due to the robustness of the metabolicnetwork and the differences between individual cancer cell lines in their metaboliccharacteristics. We take a systems biology approach. We investigate the cellular bioenergetic profiles of a panel of five non-small celllung cancer cell lines before and after treatment with a novel inhibitor of theglutaminase-1 (GLS1) enzyme. Additionally, we explore the effects of this inhibitoron intracellular metabolism of these cell lines as well as on the uptake and secretionof glucose, lactate and amino acids. To be able to do the latter robustly, wehad to modify the experimental assay considerably from procedures that seemto be standard in the literature; using these earlier procedures the metabolicenvironment of the cells was highly variable, leading to misleading results onthe metabolic effects of the inhibitor. We reduced cell density, altered mediumvolume and changed the time-window of the assay. This led to the cells growingexponentially, appearing indifferent to the few remaining changes. In this newassay, the metabolic effects of the glutaminase inhibitor became robust. One of the most significant results of this study is the metabolic heterogeneitydisplayed across the cell line panel under basal conditions. Differences in themetabolic functioning of the cell lines were observed in terms of both theirbioenergetic and metabolic profile. The amount of respiration attributed tooxidative phosphorylation differed between cell lines and respiratory capacity wasattenuated in most cells. However, the rate of glycolysis was similar betweencell lines in this assay. These results suggest that the Warburg effect arisesthrough a greater diversity of mechanisms than traditionally assumed, involvingvarious combinations of changes in the expression of glycolytic and mitochondrialmetabolic enzymes. The effects of GLS1 inhibition on cellular bioenergetics and metabolism alsodiffered between cell lines, even between resistant cell lines, indicating that theremay also be a diversity of resistance mechanisms. The metabolomic response ofcell lines to treatment suggests potential resistance mechanisms through metabolicadaptation or through the prior differences in the metabolic function of resistantcell lines. Part of the metabolome response to GLS1 inhibition was quite specificfor sensitive cells, with high concentrations of IMP as the strongest marker. Our results suggest that the metabolome is a significant player in what determinesthe response of cells to metabolic inhibitors, that its responses differ between cancercells, that responses are not beyond systems understanding, and that thereforethe metabolome should be taken into account in the design of and therapy withanti-cancer drugs.

>Magister Scientiae - MSc / New therapeutic strategies are needed for a diverse array of poorly understood neurological impairments. These include neurodegenerative disorders such as Parkinson’s disease and Alzheimer’s disease, and the psychiatric disorders such as depression, anxiety and drug dependence. Popular neuropharmacotherapies have focused on dopamine (DA), serotonin (5HT), γ-aminobutric acid (GABA) and glutamate systems (Jupp & Lawrence, 2010). However recent research points to the sigma receptor (σR) as a possible neuromodulatory system. Due to its multi-receptor action, the σR can trigger several significant biological pathways. This indicates its ideal potential as a drug target to effectively minimise drug dosage and potential side effects. Currently there are a limited number of σR ligands available and few possess the selectivity to significantly show σR’s role in neurological processes. Polycyclic amines have shown notable sigma activity and provide an advantageous scaffold for drug design that can improve pharmacodynamic and pharmacokinetic properties (Banister et al., 2010; Geldenhuys et al., 2005). Aryl-heterocycle amine groups were also shown to improve σR activity (Piergentili et al., 2009).

Deciphering the activity of chemical molecules against a pathogenic organism is an essential task in drug discovery process. Virtual screening, in which few plausible molecules are selected from a large set for further processing using computational methods, has become an integral part and complements the expensive and time-consuming in vivo and in vitro experiments. To this end, it is essential to extract certain features from molecules which in the one hand are relevant to the biological activity under consideration, and on the other are suitable for designing fast and robust algorithms. The features/representations are derived either from physicochemical properties or their structures in numerical form and are known as descriptors. In this work we develop two new molecular-fragment descriptors based on the critical analysis of existing descriptors. This development is primarily guided by the notion of coding degeneracy, and the ordering induced by the descriptor on the fragments. One of these descriptors is derived based on the simple graph representation of the molecule, and attempts to encode topological feature or the connectivity pattern in a hierarchical way without discriminating atom or bond types. Second descriptor extends the first one by weighing the atoms (vertices) in consideration with the bonding pattern, valence state and type of the atom. Further, the usefulness of these indices is tested by ranking and classifying molecules in two previously studied large heterogeneous data sets with regard to their anti-tubercular and other bacterial activity. This is achieved by developing a scoring function based on clustering using these new descriptors. Clusters are obtained by ordering the descriptors of training set molecules, and identifying the regions which are (almost) exclusively coming from active/inactive molecules. To test the activity of a new molecule, overlap of its descriptors in those cluster (interpolation) is weighted. Our results are found to be superior compared to previous studies: we obtained better classification performance by using only structural information while previous studies used both structural features and some physicochemical parameters. This makes our model simple, more interpretable and less vulnerable to statistical problems like chance correlation and over fitting. With focus on predictive modeling, we have carried out rigorous statistical validation. New descriptors utilize primarily the topological information in a hierarchical way. This can have significant implications in the design of new bioactive molecules (inverse QSAR, combinatorial library design) which is plagued by combinatorial explosion due to use of large number of descriptors. While the combinatorial generation of molecules with desirable properties is still a problem to be satisfactorily solved, our model has potential to reduce the number of degrees of freedom, thereby reducing the complexity.

Liver cancer accounts for nearly 10% of all cancers in the US. Intrahepatic Arterial Radiomicrosphere Therapy (RMT), also known as Selective Internal Radiation Treatment (SIRT), is one of the evolving treatment modalities. Successful patient clinical outcomes require suitable treatment planning followed by delivery of the microspheres for therapy. The production and in vitro evaluation of various polymers (PGCD, CHS and CHSg) microspheres for a RMT and RMT planning are described. Microparticles with a 30±10 µm size distribution were prepared by emulsion method. The in vitro half-life of the particles was determined in PBS buffer and porcine plasma and their potential application (treatment or treatment planning) established. Further, the fast degrading microspheres (≤ 48 hours in vitro half-life) were labeled with 68Ga and/or 99mTc as they are suitable for the imaging component of treatment planning, which is the primary emphasis of this dissertation. Labeling kinetics demonstrated that 68Ga-PGCD, 68Ga-CHSg and 68Ga-NOTA-CHSg can be labeled with more than 95% yield in 15 minutes; 99mTc-PGCD and 99mTc-CHSg can also be labeled with high yield within 15-30 minutes. In vitro stability after four hours was more than 90% in saline and PBS buffer for all of them. Experiments in reconstituted hemoglobin lysate were also performed. Two successful imaging (RMT planning) agents were found: 99mTc-CHSg and 68Ga-NOTA-CHSg. For the 99mTc-PGCD a successful perfusion image was obtained after 10 minutes, however the in vivo degradation was very fast (half-life), releasing the 99mTc from the lungs. Slow degrading CHS microparticles (> 21 days half-life) were modified with p-SCN-b-DOTA and labeled with 90Y for production of 90Y-DOTA-CHS. Radiochemical purity was evaluated in vitro and in vivo showing more than 90% stability after 72 and 24 hours respectively. All agents were compared to their respective gold standards (99mTc-MAA for 68Ga-NOTA-CHSg and 99mTc-CHSg; 90Y-SirTEX for 90Y-DOTA-CHS) showing superior in vivo stability. RMT and RMT planning agents (Therapy, PET and SPECT imaging) were designed and successfully evaluated in vitro and in vivo.

Mon projet a porté sur la caractérisation du mécanisme moléculaire des enzymes d'arenavirus (une 3'-5' exoribonucléase et une endonuclease) impliquées dans l'inhibition de la réponse innée IFN de type I et dans le vole de coiffe respectivement, et le développement d'une stratégie thérapeutique basée sur leur structures. Premièrement, j'ai résolu deux structures cristallographiques à haute résolution du domaine exoribonucléases du virus Mopeia (NP-exo MOPV) -un homologue du virus Lassa pathogène- en complexe avec deux ions différents. Ensuite, j'ai effectué une caractérisation fonctionnelle de l’activité exoribonucléase 3'-5' codée par ce domaine. Une corrélation entre la structure et la fonction de NP-exo MOPV démontre que; L’activité exoribonucléase 3'-5' est conservée chez les arenavirus pathogènes ainsi que chez les non-pathogènes. J'ai démontré pour la première fois que l'exoribonucléase est capable d'exciser un ARN misapparié, suggérant ainsi une potentielle activité de correction d'erreur par cette enzyme. Avec la structure de NP-exo MOPV, j'ai développé une stratégie in silico pour identifier des inhibiteurs potentiels spécifiques contre son activité et un inhibiteur a était identifié.En parallèle, nous avons résolu deux structures cristallographiques du domaine de l'endonuclease du virus de la LCMV en complexe avec deux ions catalytiques et deux composés appartenant a la famille des diketo. En résumé, ce travail éclaircit le rôle des exoribonucléases de la famille d'Arenaviridae allant de l’évasion de l'immunité innée à son implication directe dans la réplication. Il ouvre également la voie au développement des inhibiteurs contre ces nucléases. / My PhD work focused on the characterization of the molecular mechanism of two arenavirus enzymes - a 3'-5' exoribonuclease and an endonuclease - implicated in type I IFN suppression and mRNA cap-snatching respectively and the design of a structure based-drug strategy against them. First I solved two high resolution crystal structures of the exoribonuclease domain of Mopeia virus (NP-exo MOPV) -a non pathogenic homologue of the highly pathogenic Lassa virus- in complex with different metal ions. Next I performed an in depth functional characterization of the 3'-5' exoribonuclease activity encoded by this domain. By correlating the structure and function of NP-exo MOPV, I showed that; the 3'-5' exoribonuclease activity is conserved in pathogenic as well as in non-pathogenic arenaviruses. Also, I showed for the first time that this enzyme is able to excise a mismatched RNA suggesting that, arenaviruses might posses a mechanism to limit error incorporation by the RdR polymerase during replication. Using the crystal structure of NP-exo MOPV I designed a structure-based strategy to identify potential inhibitors specific for the 3'-5' exoribonuclease activity and have identified a potential inhibitor.Alongside, we solved two crystal structures of the endonuclease domain of LCMV in complex with two catalytic ions and two compounds belonging to the diketo family.In conclusion, this work has a deep implication extending the role of the Arenaviridae exoribonuclease from innate immunity evasion to direct implication in replication. It also paves the way for the development of inhibitors against these arenavirus nucleases.

Computational drug design methods have great potential in drug discovery particularly in lead identification and lead optimization. 3-Phosphoinositide dependent kinase-1 (PDK1) is a protein kinase and a well validated anti-cancer target. Inhibitors of PDK1 have the potential to be developed as anti-cancer drugs. In this work, we have applied various novel computational drug design strategies to design and identify new PDK1 inhibitors with potential anti-cancer activity. We have pursued novel structure-based drug design strategies and identified a new binding mode for celecoxib and its derivatives binding with PDK1. This new binding mode provides a valuable basis for rational design of potent PDK1 inhibitors. In order to understand the structure-activity relationship of indolinone-based PDK1 inhibitors, we have carried out a combined molecular docking and three-dimensional quantitative structure-activity relationship (3D-QSAR) modeling study. The predictive ability of the developed 3D-QSAR models were validated using an external test set of compounds. An efficient strategy of the hierarchical virtual screening with increasing complexity was pursued to identify new hits against PDK1. Our approach uses a combination of ligand-based and structure-based virtual screening including shape-based filtering, rigid docking, and flexible docking. In addition, a more sophisticated molecular dynamics/molecular mechanics- Poisson-Boltzmann surface area (MD/MM-PBSA) analysis was used as the final filter in the virtual screening. Our screening strategy has led to the identification of a new PDK1 inhibitor. The anticancer activities of this compound have been confirmed by the anticancer activity assays of national cancer institute-developmental therapeutics program (NCI-DTP) using 60 cancer cell lines. The PDK1-inhibitor binding mode determined in this study may be valuable in future de novo drug design. The virtual screening approach tested and used in this study could also be applied to lead identification in other drug discovery efforts.

Drug resistance is a global health threat costing society billions of dollars and impacting millions of lives each year. Current drug design strategies are inadequate because they focus on disrupting target activity and not restricting the evolutionary pathways to resistance. Improved strategies would exploit the structural and dynamic changes in the enzyme–inhibitor system integrating data from many inhibitors and variants. Using HIV-1 protease as a model system, I aimed to elucidate the underlying resistance mechanisms, characterize conserved protease-inhibitor interactions, and generate more robust inhibitors by applying these insights. For primary mechanisms of resistance, comparing interactions at the protease–inhibitor interface showed how specific modifications affected potency. For mutations distal to the active site, molecular dynamics simulations were necessary to elucidate how changes propagated to reduce inhibitor binding. These insights informed inhibitor design to improve potency against highly resistant variants by optimizing hydrogen bonding. A series of hybrid inhibitors was also designed that showed excellent potency by combining key moieties of multiple FDA-approved inhibitors. I characterized the structural basis for alterations in binding affinity in HIV-1 protease both from mutations and inhibitors. I applied these strategies to HTLV-1 protease, a potential drug target. I identified the HIV-1 inhibitor darunavir as a viable scaffold and evaluated analogues, leading to a low-nanomolar compound with potential for optimization. Hopefully, insights from this thesis will lead to the development of potent HTLV-1 protease inhibitors. More broadly, these inhibitor design strategies are applicable to other rapidly evolving targets, thereby reducing drug resistance rates in the future.

ASF1 est un chaperon d’histones H3-H4 impliqué dans de nombreux cancers. Comme bon nombre de protéines, ce chaperon exerce ses fonctions dans la cellule à travers des interactions protéine-protéine qu’il établit avec d’autres partenaires protéiques. La présente thèse porte sur le développement d’une stratégie originale de design de peptides inhibiteurs de ce type d’interactions souvent associées à des maladies. Cette stratégie rationnelle et itérative repose sur le couplage d’épitopes de liaison provenant de différents partenaires de l’interaction, et leur stabilisation par l’introduction de résidus « ancre » permettant ainsi d’engager un grand nombre de contacts avec la cible. L’extension de cette approche vers des peptidomimes permet par la suite de surmonter les obstacles liés à l’utilisation des peptides en thérapeutique tels que la biodisponibilité et la demi-vie. Appliquée au ciblage d’ASF1, cette méthode a permis de concevoir un peptide, ip4, présentant une affinité de 3nM pour sa cible, soit 3000 fois supérieure au partenaire naturel H3. Ce même peptide a été mimé avec succès par un composé, if3, de nature oligourée. Efficacement internalisés à l’aide d’une Cell Penetrating Peptide clivable, ces inhibiteurs présentent un effet antiprolifératif provoquant la mort des cellules cancéreuse, vraisemblablement dû au ciblage spécifique d’ASF1. / ASF1 is a histone H3-H4 chaperone implicated in several cancers. Like many proteins, this chaperone mediates its cellular functions through protein-protein interactions involving various protein partners. The present thesis focuses on the development of an original strategy to design inhibitory peptides targeting such disease-associated type of biological interactions. This rational and iterative strategy relies on the tethering of binding epitopes isolated from different partners, and stabilized by “anchor” residues that engage large number of atomic contacts with the target. The further progression of this approach toward a peptidomimetic strategy overcomes obstacles commonly associated to the therapeutic use of peptides such as biodisponibility and half-life. Applied for targeting ASF1, such method allowed the conception of a peptide, ip4, presenting a 3nM affinity for its target, which is 3000 fold higher than that of the natural partner H3. This peptide could be successfully mimicked by an oligourea structure, giving rise to the peptidomimetic if3. When coupled to a cleavable Cell Penetrating Peptide, these inhibitors displayed an on-target effect where they impeded cancerous cells proliferation, ultimately resulting in cells death.

A number of cyclic peptides containing a positively charged ring composed of arginine residues attached to hydrophobic tail made of tryptophan residues through a lysine linker namely [R5K]W5, [R6K]W5, [R5K]W6, [R7K]W5, [R5K]W7, [R6K]W6, and [R7K]W7 were synthesized and evaluated as molecular transporters. The peptides were evaluated for their ability to deliver, fluorescence-labeled cell-impermeable negatively charged phosphopeptide (F′-GpYEEI), and fluorescent labeled anti-HIV drugs (F′-FTC and F′-d4T). The results indicated that the presence of positively charged arginine residues on the ring and hydrophobic tryptophan residues in a sequential linear outside the ring was an optimal approach to improve the intracellular uptake of cargo molecules through non-covalent interactions. Some of these peptides were also evaluated for their efficiency for intracellular delivery of siRNA to triple-negative breast cancer cell lines in the presence and absence of 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE). [R6K]W6 and [R5K]W5 were found to be very efficient in the delivery of siRNA. Furthermore, co-formulation of peptides with lipid DOPE significantly enhanced the efficiency of siRNA delivery compared to peptide alone. Silencing of kinesin spindle protein (KSP) and Janus kinase 2 (JAK2) was evaluated in MDA-MB-231 cells in the presence of the peptides. The addition of DOPE significantly enhanced the silencing efficiency for all selected peptides. A chemotherapeutic drug, doxorubicin (Dox) was covalently conjugated to the cyclic peptide [R5K]W7A and linear peptide R5KW7A, and the biological activity was evaluated in cell-based assays. Comparative antiproliferative assays between covalently conjugated peptide-Dox and the corresponding noncovalent physical mixtures of the peptides and Dox were performed. The conjugation of Dox with cyclic [R5K]W7A-Dox exhibited similar antiproliferative activity compared to Dox alone after 72 h incubation time in all cancer cell lines, such as leukemia, ovarian and gastric cancer cells. However, [R5K]W7A-Dox significantly reduced the cell cytotoxicity in normal cell lines such as normal heart muscle and normal kidney cells after 72 h when compared with Dox alone. These results revealed that this cyclic peptide prodrug can be used as a potential candidate for the treatment of cancer cells with reduced side effects against normal cells in the body.