Teses / dissertações sobre o tema "Structure-Based approaches"

This dissertation thesis consists of a series of chapters that are interwoven by solving interesting biological problems, employing various computational methodologies. These techniques provide meaningful physical insights to promote the scientific fields of interest. Focus of chapter 1 concerns, the importance of computational tools like docking studies in advancing structure based drug design processes. This chapter also addresses the prime concerns like scoring functions, sampling algorithms and flexible docking studies that hamper the docking successes. Information about the different kinds of flexible dockings in terms of accuracy, time limitations and success studies are presented. Later the importance of Induced fit docking studies was explained in comparison to traditional MD simulations to predict the absolute binding modes. Chapter 2 and 3 focuses on understanding, how sickle cell disease progresses through the production of sickled hemoglobin and its effects on sickle cell patients. And how, hydroxyurea, the only FDA approved treatment of sickle cell disease acts to subside sickle cell effects. It is believed the primary mechanism of action is associated with the pharmacological elevation of nitric oxide in the blood, however, the exact details of this mechanism is still unclear. HU interacts with oxy and deoxyHb resulting in slow NO production rates. However, this did not correlate with the observed increase of NO concentrations in patients undergoing HU therapy. The discrepancy can be attributed to the interaction of HU competing with other heme based enzymes such as catalase and peroxidases. In these two chapters, we investigate the atomic level details of this process using a combination of flexible-ligand / flexible-receptor virtual screening (i.e. induced fit docking, IFD) coupled with energetic analysis that decomposes interaction energies at the atomic level. Using these tools we were able to elucidate the previously unknown substrate binding modes of a series of hydroxyurea analogs to human hemoglobin, catalase and the concomitant structural changes of the enzymes. Our results are consistent with kinetic and EPR measurements of hydroxyurea-hemoglobin reactions and a full mechanism is proposed that offers new insights into possibly improving substrate binding and/or reactivity. Finally in chapter 4, we have developed a 3D bioactive structure of O6-alkylguanine-DNA alkyltransferase (AGT), a DNA repair protein using Monte Carlo conformational search process. It is known that AGT prevents DNA damage, mutations and apoptosis arising from alkylated guanines. Various Benzyl guanine analouges of O6- methylguanine were tested for activity as potential inhibitors. The nature and position of the substitutions methyl and aminomethyl profoundly affected their activity. Molecular modeling of their interactions with alkyltransferase provided a molecular explanation for these results. The square of the correlation coefficient (R2 ) obtained between E-model scores (obtained from GLIDE XP/QPLD docking calculations) vs log(ED)values via a linear regression analysis was 0.96. The models indicate that the ortho-substitution causes a steric clash interfering with binding, whereas the meta-aminomethyl substitution allows an interaction of the amino group to generate an additional hydrogen bond with the protein. Using this model for virtually screening studies resulted in identification of seven lead compounds with novel scaffolds from National Cancer Institute Diversity Set2.

Les technologies logicielles ne cessent d'évoluer pour faciliter le développement, le déploiement et la maintenance d'applications dans différents domaines. En parallèle, ces applications évoluent en continu pour garantir une bonne qualité de service et deviennent de plus en plus complexes. Cette évolution implique souvent des coûts de développement et de maintenance de plus en plus importants, auxquels peut s'ajouter une augmentation des coûts de déploiement sur des infrastructures d'exécution récentes comme le cloud. Réduire ces coûts et améliorer la qualité de ces applications sont actuellement des objectifs centraux du domaine du génie logiciel. Récemment, les microservices sont apparus comme un exemple de technologie ou style architectural favorisant l'atteinte de ces objectifs.Alors que les microservices peuvent être utilisés pour développer de nouvelles applications, il existe des applications monolithiques (i.e., monolithes) cons-truites comme une seule unité et que les propriétaires (e.g., entreprise, etc.) souhaitent maintenir et déployer sur le cloud. Dans ce cas, il est fréquent d'envisager de redévelopper ces applications à partir de rien ou d'envisager une migration vers de nouveaux styles architecturaux. Redévelopper une application ou réaliser une migration manuellement peut devenir rapidement une tâche longue, source d'erreurs et très coûteuse. Une migration automatique apparaît donc comme une solution évidente.L'objectif principal de notre thèse est de contribuer à proposer des solutions pour l'automatisation du processus de migration d'applications monolithiques orientées objet vers des microservices. Cette migration implique deux étapes : l'identification de microservices et le packaging de ces microservices. Nous nous focalisons sur d'identification en s'appuyant sur une analyse du code source. Nous proposons en particulier deux approches.La première consiste à identifier des microservices en analysant les relations structurelles entre les classes du code source ainsi que les accès aux données persistantes. Dans cette approche, nous prenons aussi en compte les recommandations d'un architecte logiciel. L'originalité de ce travail peut être vue sous trois aspects. Tout d'abord, les microservices sont identifiés en se basant sur l'évaluation d'une fonction bien définie mesurant leur qualité. Cette fonction repose sur des métriques reflétant la "sémantique" du concept "microservice". Deuxièmement, les recommandations de l'architecte logiciel ne sont exploitées que lorsqu'elles sont disponibles. Enfin, deux modèles algorithmiques ont été utilisés pour partitionner les classes d'une application orientée objet en microservices : un algorithme de regroupement hiérarchique et un algorithme génétique.La deuxième approche consiste à extraire à partir d'un code source orienté objet un workflow qui peut être utilisé en entrée de certaines approches existantes d'identification des microservices. Un workflow décrit le séquencement de tâches constituant une application suivant deux formalismes: un flot de contrôle et/ou un flot de données. L'extraction d'un workflow à partir d'un code source nécessite d'être capable de définir une correspondance entre les concepts du mon-de objet et ceux d'un workflow.Pour valider nos deux approches, nous avons implémenté deux prototypes et mené des expérimentations sur plusieurs cas d'étude. Les microservices identifiés ont été évalués qualitativement et quantitativement. Les workflows obtenus ont été évalués manuellement sur un jeu de tests. Les résultats obtenus montrent respectivement la pertinence des microservices identifiés et l'exactitude des workflows obtenus
Software technologies are constantly evolving to facilitate the development, deployment, and maintenance of applications in different areas. In parallel, these applications evolve continuously to guarantee an adequate quality of service, and they become more and more complex. Such evolution often involves increased development and maintenance costs, that can become even higher when these applications are deployed in recent execution infrastructures such as the cloud. Nowadays, reducing these costs and improving the quality of applications are main objectives of software engineering. Recently, microservices have emerged as an example of a technology or architectural style that helps to achieve these objectives.While microservices can be used to develop new applications, there are monolithic ones (i.e., monoliths) built as a single unit and their owners (e.g., companies, etc.) want to maintain and deploy them in the cloud. In this case, it is common to consider rewriting these applications from scratch or migrating them towards recent architectural styles. Rewriting an application or migrating it manually can quickly become a long, error-prone, and expensive task. An automatic migration appears as an evident solution.The ultimate aim of our dissertation is contributing to automate the migration of monolithic Object-Oriented (OO) applications to microservices. This migration consists of two steps: microservice identification and microservice packaging. We focus on microservice identification based on source code analysis. Specifically, we propose two approaches.The first one identifies microservices from the source code of a monolithic OO application relying on code structure, data accesses, and software architect recommendations. The originality of our approach can be viewed from three aspects. Firstly, microservices are identified based on the evaluation of a well-defined function measuring their quality. This function relies on metrics reflecting the "semantics" of the concept "microservice". Secondly, software architect recommendations are exploited only when they are available. Finally, two algorithmic models have been used to partition the classes of an OO application into microservices: clustering and genetic algorithms.The second approach extracts from an OO source code a workflow that can be used as an input of some existing microservice identification approaches. A workflow describes the sequencing of tasks constituting an application according to two formalisms: control flow and /or data flow. Extracting a workflow from source code requires the ability to map OO conceptsinto workflow ones.To validate both approaches, we implemented two prototypes and conducted experiments on several case studies. The identified microservices have been evaluated qualitatively and quantitatively. The extracted workflows have been manually evaluated relying on test suites. The obtained results show respectively the relevance of the identified microservices and the correctness of the extracted workflows

Computer Aided Drug Design/Discovery methods became complementary to traditional and modern drug discovery approaches. Indeed CADD is useful to improve and speed up the detection and the optimization of bioactive molecules. The present study is focused on the application of structure-based approaches to the study of pharmaceutical relevant targets. The introduction provides a quick overview on the fundamentals of computational chemistry and structure-based methods, while in the successive chapters the main targets investigated through these methods are treated. In particular we focused our attention on Reverse Transcriptase of HIV-1, Monoamine oxidase B and VP35 of Ebola virus. The last chapter is dedicated to the validation of covalent docking performed with Autodock.

Pharmaceutical Sciences
Ph.D.
In drug discovery there are several approaches to lead generation and one traditional approach involves the synthesis and screening of a structurally diverse compound library against a number of biological targets to identify high affinity lead compounds. The use of a `privileged' structure-based compound library represents a viable approach that could lead to drug like lead compounds. Privileged structures are defined as those ligand substructures that may be used to generate high affinity leads for more than one type of receptor. Examples of privileged structures include phenyl substituted monocycles such as biphenyls, diphenyl methane derivatives, 1,4-dihydropyridines, fused ring systems such as chromones, quinoxalines, quinazolines, 2-benzoxazolones, indoles, benzimidazoles and benzofurans. There are several instances in the literature describing the development of compound libraries based on privileged structures with reportedly high hit rates. Privileged structure based approaches has been used with notable success in the identification of high affinity ligands especially for G-protein coupled receptors (GPCRs). The scaffold 2-aminothiazole (fused and non-fused) may be considered a privileged structure because of its occurrence in a wide variety of pharmaceuticals. The scaffold is found in antibacterials, anti-inflammatory agents, glutamate transporter (GLT-1) modulators, serotonin and muscarinic ligands. The present study involves the synthesis of a 2-aminothiazole (fused and non-fused) based compound library (60 compounds) by incorporating bioactive fragments shown to produce hits in the biological targets of interest. Microwave assisted organic synthesis (MAOS) has been employed at key steps of scaffold synthesis as well as in Suzuki coupling to generate the target aminothiazoles. Preliminary biological screening has resulted in the identification of some promising lead compounds. Trifluoromethoxy substituted aminothiazoles were found to be potent antimicrobials with MIC values in the range of 4-16 microgram/ml. Furanone based aminothiazoles showed affinity for muscarinic receptors. Piperidine based aminothiazoles showed greater than 90% of control (8-OH-DPAT) specific agonist response at the 5-HT1A receptor subtype. The Clog P values of the most potent antimicrobials were found to be in the range of 4.5-6.2 indicating the high lipophilicity of the compounds. High lipophilicity is known to cause solubility issues that may hamper future development. Therefore in an effort to make compounds with intermediate lipophilicity, the phenyl core of the potent aminothiazoles will be replaced with pyridine core using literature procedures (Pyridine core containing aminothiazoles showed Clog P < 4). Future plans include expanding the library, improving the yields of compounds and to evaluate the compounds as modulators of glutamate transporter (GLT-1). The work could be extended to include other privileged structures such as 2-aminooxazole, 2-aminobenzoxazole, 2-aminoimidazole and 2-aminobenzimidazole. These mono and bicyclic heterocyles may be considered bioisosteres of 2-aminothiazole.
Temple University--Theses

Computer simulations of soft matter require a compromise to be made between the computational efficiency and the resolution of a model studied. Highly resolved models can give insights into the interactions between individual atoms in a soft material. But, since these atomistic models are computational expensive, they are limited to small length scales and short time scales, which makes it difficult to compare simulation results with the ones from experiments in the laboratory. On the other hand, continuum models enable the study of soft matter at larger length and longer time scales and are computationally less expensive, but they rather focus on macroscopic properties than on their atomistic origin. A possible way to bridge the gap between these scales is to perform simulations at an intermediate level of resolution. The problem, which exists at this mesoscopic scale, is the lack of accurate models. Hence, new ones have to be built. The process to construct mesoscopic models based on information from the atomistic scale is commonly referred to as bottom-up coarse graining. Bottom-up coarse graining describes the process of lowering the resolution of a atomistic model to make it applicable at larger length and time scales. The major goal of this Ph.D. thesis is to increase the knowledge on so called structure-based bottom-up coarse graining techniques.These methods enable the derivation of coarse grained (CG) models, which accurately reproduce the structure of an atomistic or fine grained (FG) model at the mesoscopic scale. The shortcomings of different structure-based methods are carefully analyzed and new approaches to overcome them are presented.

Les interactions protéine-protéine (IPP) jouent un rôle crucial dans de nombreuses voies biologiques et sont de plus en plus explorées en tant que cibles thérapeutiques potentielles, notamment pour le traitement des maladies infectieuses. Cependant, la conception de petites molécules modulatrices pour les IPP reste un défi, car les interfaces des IPP n'ont pas évolué pour lier des petites molécules comme les cibles thérapeutiques conventionnelles telles que les enzymes ou les récepteurs membranaires. Par conséquent, la preuve de leur drugabilité doit être apportée au cas par cas. Dans ce contexte, les approches computationnelles peuvent être utiles pour aider à la conception de modulateurs IPP. Ce travail vise à développer de nouveaux protocoles de conception de médicaments in silico spécifiquement adaptés aux cibles IPP, dans le but de concevoir de nouveaux médicaments antiviraux contre deux cibles IPP : le virus respiratoire syncytial (VRS) et le SARS-CoV-2
Protein-Protein interactions (PPI) play crucial roles in many biological pathways and are being increasingly explored as potential therapeutic targets, including for treating infectious diseases. However, designing small molecule modulators for PPI remains challenging as PPI interfaces have not evolved to bind small molecules like conventional drug targets such as enzymes or membrane receptors. Therefore, proof of their druggability must be made on a case-by-case basis. In this context, computational approaches can be useful in assisting the design of PPI modulators.This work aims to develop new in silico drug design protocols specifically tailored to PPI targets, with the goal of designing new antiviral drugs against two PPI targets: the respiratory syncytial virus (RSV) and the SARS-CoV-2

The research described in this dissertation is focused on the knowledge-based, often in silico assisted design, targeted synthesis, and biological evaluation of small molecules of interest for two translational medicinal chemistry projects. The first project (Part 1) is aimed at the identification of blood brain barrier (BBB) penetrable dual specificity tyrosine phosphorylation regulated kinase-1A (DYRK1A) inhibitors as a potential disease modifying approach to mitigate cognitive deficits associated with Alzheimer's neurodegeneration. Two major series with potent activity against DYRK1A were identified in addition to a number of other chemotype sub-series that also exhibit somewhat promising activity. Extensive profiling of active analogs revealed interesting biological activity and selectivity, which led to the identification of two analogs for in vivo studies and revealed new opportunities for further investigation into other kinase targets implicated in neurodegeneration and polypharmacological approaches. The second project (Part 2) is focused on the development of compounds that inhibit PGE₂ production, while not affecting cyclooxygenase (COX) activity, as a novel approach to treat cancer. Compounds were designed with the intention of inhibiting microsomal prostaglandin E₂ synthase-1 (mPGES-1); however, biological evaluation revealed phenotypically active compounds in a cell based assay with an unknown mechanism of action. Further profiling revealed promising anticancer activity in xenograft mouse models. In addition, PGE₂ has been implicated in an immune evasion mechanism of F. tularensis, a strain of bacteria that remains an exploitable threat in biowarfare, thus a small number of analogs were evaluated in a cell model of F. tularensis infection stimulated PGE₂ production.

The discovery and commercialization of a new drug is a long and expensive process. Such process is divided into different phases during which the phisico-chemical and therapeutic properties of the compounds are determined. In particular, the aim of the first phase is to verify whether the compound recognises and interacts efficiently with the target protein. In the last decade, several computational tools have been developed and used to support experimentalists. For this purpose, the scientist have to deal with high complex systems that are difficult to study in whole; thus, the methods and algorithms developers have to strongly simplify the system treatment. Moreover, the time required to obtain the results depends on the computational resources (hardware) available. Fortunately, the technological progress have increased the computing power at low cost, resulting in new and more complex techniques development. During this Ph.D. project we were focused on the development and even the improvement of in silico methods, which allowed to answer certain questions by saving time and money. Furthermore, these methods were implemented in software presenting a Graphical Unit Interface (GUI) with the aim to enhance the user-friendliness. The computational techniques often require a high understanding of the methodology theoretical aspects and also a good informatics proficiency, like different type files handling and hardware management. For this reason, our developed software were organized as pipelines to automatize the entire process and to make this tools useful also for non-expert users. Finally, these methodologies were applied in several research projects demonstrating their usefulness by elucidating, for the first time, interesting aspects of the ligand-protein recognition pathway.
La scoperta e la commercializzazione di un nuovo farmaco è un processo lungo e dispendioso, che si articola in diverse fasi durante le quali vengono determinate le proprietà fisiche, chimiche e terapeutiche dei composti investigati. In particolare, nella prima fase di questo processo si cerca di verificare che il composto riconosca e interagisca efficacemente con la proteina bersaglio. A tale scopo, negli ultimi decenni numerosi strumenti computazionali sono stati sviluppati e utilizzati per supportare i ricercatori che si adoperano nella parte sperimentale. I problemi affrontati presentano un alto livello di complessità, che sarebbero difficili da studiare in toto, perciò gli sviluppatori di metodi e algoritmi devono necessariamente adottare notevoli semplificazioni. Inoltre, le risorse di calcolo (hardware) determinano le tempistiche con le quali è possibile ottenere il risultato richiesto. In tal senso, lo sviluppo tecnologico ha portato a un importante aumento della potenza di calcolo a costi accessibili, stimolando l’interesse per lo sviluppo di tecniche sempre più complesse. Durante questo progetto di dottorato ci si è focalizzati sullo sviluppo e il miglioramento di metodi in silico, che permettono di rispondere ad alcuni interrogativei a costi e tempistiche di molto ridotte. Inoltre, tali metodi sono stati implementati in software dotati di interfaccia grafica (GUI) al fine di poter aiutare l’utente nel loro utilizzo. Le tecniche computazionali spesso richiedono un’elevata conoscenza teorica delle metodologie e anche una certa competenza informatica, come la gestione di diversi tipologie di file e delle risorse hardware da impiegare. Per questo motivo i software da noi sviluppati sono stati organizzati in pipelines, in modo da automatizzare l’intero processo e rendere questi strumenti fruibili anhce a persone non esperte. Infine, l’utilità di queste nuove metodologie è stata comprovata in progetti in cui questi strumenti hanno permesso di delucidare aspetti interessanti e fino ad ora non ancora accessibili nell’ambito del riconoscimento proteina-ligando.

Determining the correct state of a protein or a protein complex is of paramount importance for current medical and pharmaceutical research. The stable conformation of such systems depend on two processes called protein folding and protein-protein interaction. In the course of the last 50 years, both processes have been fruitfully studied. Yet, a complete understanding is still not reached, and the accuracy and the efficiency of the approaches for studying these problems is not yet optimal. This thesis is devoted to devising physical and statistical methods for recognizing the native state of a protein or a protein complex. The studies will be mostly based on BACH, a knowledge-based potential originally designed for the discrimination of native structures in protein folding problems. BACH method will be analyzed and extended: first, a new method to account for protein-solvent interaction will be presented. Then, we will describe an extension of BACH aimed at assessing the quality of protein complexes in protein-protein interaction problems. Finally, we will present a procedure aimed at predicting the structure of a complex based on a hierarchy of approaches ranging from rigid docking up to molecular dynamics in explicit solvent. The reliability of the approaches we propose will be always benchmarked against a selection of other state-of-the-art scoring functions which obtained good results in CASP and CAPRI competitions.

Bioinformatics is an interdisciplinary field that develops methods and software tools for understanding biological data, in particular when the data sets are large and complex. As an interdisciplinary field of science, bioinformatics combines biology, computer science, information engineering, mathematics and statistics to analyze and interpret the biological data. Bioinformatics has been used for in silico analyzes of biological queries using mathematical and statistical techniques. During the three years of doctorate we focused about Structural Bioinformatics and in particular in studies that allow the research of potential inhibitors and allosteric modulators, both against soluble and membrane proteins , related in particular with cancer and infectious deseases. All the projects were automated thanks to Python and Bash programming languages. The main operative system was Linux. The projects results are very promising and the some of those are still in progress

Nrf2 project: The protein nuclear factor erythroid 2-related factor 2 (Nrf2) is a transcription factor that provides protection against oxidative stress and the dysfunction of this pathway has been suggested to be implicated in many neurodegenerative diseases. The aim of this thesis was to identify novel Nrf2 activators that disrupt the protein-protein interaction between Nrf2 and Keap1 and thereby induce increased expression of antioxidant enzymes and protective genes. The crystal structure of the Keap1-Nrf2 interface was used to perform a virtual screen and compounds from the screen were assayed using a cellular nuclear complementation assay that measures the nuclear translocation of Nrf2 from the cytosol. Although two novel compounds were found to increase the Nrf2 nuclear translocation, they had low activity and further characterisation did not provide sufficient evidence of a Nrf2-Keap1 robust interaction. iGluRs project: AMPA and kainate receptors are ionotropic glutamate receptors (iGluRs) that are important for excitatory transmission and synaptic plasticity and are linked to several neurological disorders such as epilepsy, schizophrenia and autism. This project aimed to find novel allosteric modulators binding in the ligand-binding domain (LBD) of the GluA2 and GluK1 and GluK2 subtypes of AMPA and kainate receptors, respectively, using protein purification and X-ray crystallography methodologies. Fragment screening for GluA2 identified eight novel fragments, five of which were located at the dimer interface and three located in a novel site near the glycine-threonine dipeptide linker. As regards kainate receptors, structural information on the Gluk1 and GluK2 LBD was obtained, both proteins were soaked with in-house fragments with one compound displaying 20% occupancy in the GluK2 dimer interface. These data form the basis of future studies in the search for novel drugs for the treatment of epilepsy and schizophrenia.

The research described in this PhD thesis focuses on proteomics approaches to study the effect of oxidation on the modification status and protein-protein interactions of PTEN, a redox-sensitive phosphatase involved in a number of cellular processes including metabolism, apoptosis, cell proliferation, and survival. While direct evidence of a redox regulation of PTEN and its downstream signaling has been reported, the effect of cellular oxidative stress or direct PTEN oxidation on PTEN structure and interactome is still poorly defined. In a first study, GST-tagged PTEN was directly oxidized over a range of hypochlorous acid (HOCl) concentration, assayed for phosphatase activity, and oxidative post-translational modifications (oxPTMs) were quantified using LC-MS/MS-based label-free methods. In a second study, GSTtagged PTEN was prepared in a reduced and reversibly H2O2-oxidized form, immobilized on a resin support and incubated with HCT116 cell lysate to capture PTEN interacting proteins, which were analyzed by LC-MS/MS and comparatively quantified using label-free methods. In parallel experiments, HCT116 cells transfected with a GFP-tagged PTEN were treated with H2O2 and PTENinteracting proteins immunoprecipitated using standard methods. Several high abundance HOCl-induced oxPTMs were mapped, including those taking place at amino acids known to be important for PTEN phosphatase activity and protein-protein interactions, such as Met35, Tyr155, Tyr240 and Tyr315. A PTEN redox interactome was also characterized, which identified a number of PTEN-interacting proteins that vary with the reversible inactivation of PTEN caused by H2O2 oxidation. These included new PTEN interactors as well as the redox proteins peroxiredoxin-1 (Prdx1) and thioredoxin (Trx), which are known to be involved in the recycling of PTEN active site following H2O2-induced reversible inactivation. The results suggest that the oxidative modification of PTEN causes functional alterations in PTEN structure and interactome, with fundamental implications for the PTEN signaling role in many cellular processes, such as those involved in the pathophysiology of disease and ageing.

Bioinformatics applications can address the transfer of information at several stages of the central dogma of molecular biology, including transcription and translation. This dissertation focuses on using Bayesian models to interpret biological data in bioinformatics, using Markov chain Monte Carlo (MCMC) for the inference method. First, we use our approach to interpret data at the transcription level. We propose a two-level hierarchical Bayesian model for variable selection on cDNA Microarray data. cDNA Microarray quantifies mRNA levels of a gene simultaneously so has thousands of genes in one sample. By observing the expression patterns of genes under various treatment conditions, important clues about gene function can be obtained. We consider a multivariate Bayesian regression model and assign priors that favor sparseness in terms of number of variables (genes) used. We introduce the use of different priors to promote different degrees of sparseness using a unified two-level hierarchical Bayesian model. Second, we apply our method to a problem related to the translation level. We develop hidden Markov models to model linker/non-linker sequence regions in a protein sequence. We use a linker index to exploit differences in amino acid composition between regions from sequence information alone. A goal of protein structure prediction is to take an amino acid sequence (represented as a sequence of letters) and predict its tertiary structure. The identification of linker regions in a protein sequence is valuable in predicting the three-dimensional structure. Because of the complexities of both models encountered in practice, we employ the Markov chain Monte Carlo method (MCMC), particularly Gibbs sampling (Gelfand and Smith, 1990) for the inference of the parameter estimation.

Les molécules d'ARN sont devenues des cibles thérapeutiques majeures, et le ciblage par petites molécules se révèle particulièrement prometteur. Cependant, malgré leur potentiel, le domaine est encore en développement, avec un nombre limité de médicaments spécifiquement conçus pour l'ARN. La flexibilité intrinsèque de l'ARN, bien qu'elle constitue un obstacle, introduit des opportunités thérapeutiques que les outils computationnels actuels ne parviennent pas pleinement à exploiter malgré leur prédisposition. Le projet de cette thèse est de construire un cadre computationnel plus complet pour la conception rationnelle de composés ciblant l'ARN. La première étape pour toute approche structure-based est l'analyse des connaissances structurales disponibles. Cependant, il manquait une base de données complète, organisée et régulièrement mise à jour pour la communauté scientifique. Pour combler cette lacune, j'ai créé HARIBOSS, une base de données de toutes les structures expérimentalement déterminées des complexes ARN-petites molécules extraites de la base de données PDB. Chaque entrée de HARIBOSS, accessible via une interface web dédiée (https://hariboss.pasteur.cloud), est annotée avec les propriétés physico-chimiques des ligands et des poches d'ARN. Cette base de données constamment mise à jour facilitera l'exploration des composés drug-like liées à l'ARN, l'analyse des propriétés des ligands et des poches, et en fin de compte, le développement de stratégies in silico pour identifier des petites molécules ciblant l'ARN. Lors de sa sortie, il a été possible de souligner que la majorité des poches de liaison à l'ARN ne conviennent pas aux interactions avec des molécules drug-like. Cela est dû à une hydrophobicité moindre et une exposition au solvant accrue par rapport aux sites de liaison des protéines. Cependant, cela résulte d'une représentation statique de l'ARN, qui peut ne pas capturer pleinement les mécanismes d'interaction avec de petites molécules. Il était nécessaire d'introduire des techniques computationnelles avancées pour une prise en compte efficace de la flexibilité de l'ARN. Dans cette direction, j'ai mis en œuvre SHAMAN, une technique computationnelle pour identifier les sites de liaison potentiels des petites molécules dans les ensembles structuraux d'ARN. SHAMAN permet d'explorer le paysage conformationnel de l'ARN cible par des simulations de dynamique moléculaire atomistique. Dans le même temps, il identifie efficacement les poches d'ARN en utilisant de petits fragments dont l'exploration de la surface de l'ARN est accélérée par des techniques d'enhanced sampling. Dans un ensemble de données comprenant divers riboswitches structurés ainsi que de petits ARN viraux flexibles, SHAMAN a précisément localisé des poches résolues expérimentalement, les classant les régions d’interaction préférées. Notamment, la précision de SHAMAN est supérieure à celle d'autres outils travaillant sur des structures statiques d'ARN dans un scénario réaliste de découverte de médicaments où seules les structures apo de la cible sont disponibles. Cela confirme que SHAMAN est une plateforme robuste pour les futures initiatives de conception de médicaments ciblant l'ARN avec de petites molécules, en particulier compte tenu de sa pertinence potentielle dans les campagnes de criblage virtuel. Dans l'ensemble, ma recherche contribue à améliorer notre compréhension et notre utilisation de l'ARN en tant que cible pour les médicaments à petites molécules, ouvrant la voie à des stratégies thérapeutiques plus efficaces dans ce domaine en évolution
RNA molecules have recently gained huge relevance as therapeutic targets. The direct targeting of RNA with small molecule drugs emerges for its wide applicability to different classes of RNAs. Despite this potential, the field is still in its infancy and the number of available RNA-targeted drugs remains limited. A major challenge is constituted by the highly flexible and elusive nature of the RNA targets. Nonetheless, RNA flexibility also presents unique opportunities that could be leveraged to enhance the efficacy and selectivity of newly designed therapeutic agents. To this end, computer-aided drug design techniques emerge as a natural and comprehensive approach. However, existing tools do not fully account for the flexibility of the RNA. The project of this PhD work aims to build a computational framework toward the rational design of compounds targeting RNA. The first essential step for any structure-based approach is the analysis of the available structural knowledge. However, a comprehensive, curated, and regularly updated repository for the scientific community was lacking. To fill this gap, I curated the creation of HARIBOSS ("Harnessing RIBOnucleic acid - Small molecule Structures"), a database of all the experimentally-determined structures of RNA-small molecule complexes retrieved from the PDB database. HARIBOSS is available via a dedicated web interface (https://hariboss.pasteur.cloud), and is regularly updated with all the structures resolved by X-ray, NMR, and cryo-EM, in which ligands with drug-like properties interact with RNA molecules. Each HARIBOSS entry is annotated with physico-chemical properties of ligands and RNA pockets. HARIBOSS repository, constantly updated, will facilitate the exploration of drug-like compounds known to bind RNA, the analysis of ligands and pockets properties and, ultimately, the development of in silico strategies to identify RNA-targeting small molecules. In coincidence of its release, it was possible to highlight that the majority of RNA binding pockets are unsuitable for interactions with drug-like molecules, attributed to the lower hydrophobicity and increased solvent exposure compared to protein binding sites. However, this emerges from a static depiction of RNA, which may not fully capture their interaction mechanisms with small molecules. In a broader perspective, it was necessary to introduce more advanced computational techniques for an effective accounting of RNA flexibility in the characterization of potential binding sites. In this direction, I implemented SHAMAN, a computational technique to identify potential small-molecule binding sites in RNA structural ensembles. SHAMAN enables the exploration of the target RNA conformational landscape through atomistic molecular dynamics. Simultaneously, it efficiently identifies RNA pockets using small probe compounds whose exploration of the RNA surface is accelerated by enhanced-sampling techniques. In a benchmark encompassing diverse large, structured riboswitches as well as small, flexible viral RNAs, SHAMAN accurately located experimentally resolved pockets, ranking them as preferred probe hotspots. Notably, SHAMAN accuracy was superior to other tools working on static RNA structures in the realistic drug discovery scenario where only apo structures of the target are available. This establishes SHAMAN as a robust platform for future drug design endeavors targeting RNA with small molecules, especially considering its potential applicability in virtual screening campaigns. Overall, my research contributed to enhance our understanding and utilization of RNA as a target for small molecule drugs, paving the way for more effective drug design strategies in this evolving field

INTRODUCTION Meta-analysis is a powerful tool to cumulate and summarize the knowledge in a research field through statistical instruments, and to identify the overall measure of a treatment’s effect by combining several study-specific results. However, it is a controversial tool, because even small violations of certain rules can lead to misleading conclusions. Pooling data through meta-analysis can create problems, such as non linear correlations, multifactorial rather than unifactorial effects, limited coverage, or inhomogeneous data that fails to connect with the hypothesis. When head-to-head treatment comparisons are not available or conclusive, the limitations of standard (i.e. pairwise) meta-analyses can be overcome by network meta-analyses (NMA) which can provide estimates of treatment efficacy or safety of multiple treatment regimens. Different treatment strategies are analyzed by statistical inference methods rather than simply summing up trials that evaluated the same intervention compared to another intervention, standard care, or placebo. If a first trial compares drug A to drug B, showing that drug A is significantly superior to drug B, and a second trial investigates the same or a similar patient population comparing drug B versus drug C (demonstrating that drug B is equivalent to drug C), NMA may allow to infer that drug A is also potentially superior to drug C for this given patient population, even though there was no direct test of drug A against drug C. CONTENTS In this thesis we provided and discussed methods to overcome the limits of standard (univariate) meta-analysis, focusing on the ability to cope with multiple treatments and to deal with correlated data where correlation can derive from multiple endpoints, time-varying responses or from clustered observation. In the first chapter we explore the principal steps (from writing a prospective protocol of analysis to results’ interpretation) in order to minimize the risk of conducting a mediocre meta-analysis and to support researchers to accurately evaluate the published findings. The second chapter represents an overview of conceptual and practical issues of a network meta-analysis. We start from general considerations on network meta-analysis to specifically appraise how to collect study data, structure the analytical network, and specify the requirements for different models and parameter interpretations. Specifically, we outline the key steps, from literature search to sensitivity analysis, necessary to perform a valid network meta-analysis on binomial data. In the third party of this work, we focus our attention on data which can be analyzed with a binomial model applying the Bayesian hierarchical approach and using Markov Chain Monte Carlo approach. We also apply this analytical approach to a case study on the beneficial effects of anesthetic agents in order to further clarify the statistical details of the models, diagnostics, and computations. We presented a practical guide with the actual WinBUGS and SAS codes to allow transparency and ease of replication of all steps that are required when carrying out such quantitative syntheses. In the fourth chapter we propose an alternative frequentist approach to estimate consistency and inconsistency models for a network meta-analysis. We discuss the multilevel network meta-analysis which include a three-level data structure: subjects within studies at the first level, studies within study designs at the second level and design configuration at the third level. We discuss multilevel modeling which may be carried out within widely available statistical programs such as SAS software, and we compare the results of a published Bayesian network meta-analysis on a binary endpoint which examines the effect on mortality of desflurane, isoflurane, sevoflurane, and total intravenous anaesthetics at the longest follow-up available. In the final chapter we compare the Bayesian and the novel frequentist-multilevel approach in performing network meta-analysis on publicly available data and we investigate the descriptive characteristics that may contribute to decrease or increase the potential difference between the estimates derived from the two approaches. The two approaches were compared in terms of the difference between the pooled estimates or their standardized values, and of the Euclidean distance. BAYESIAN NETWORK META-ANALYSIS Suppose that J trials provide mixed comparisons among K treatments and that a is the trial-specific reference treatment. The random effect model is defined by: yja= β0+eja for j=1,2,...,J; a=1,2,…,K-1 yjk= β0j+δj,ak+ejk for j=1,2,...,J; a=2,3,…,K; b