La compréhension de la relation structure-propriétés des parois des cellules végétales s'appuie de plus en plus sur l'utilisation d'approches de modélisation moléculaire en général et de dynamique moléculaire en particulier. A ce jour, le poids numérique que représente une telle démarche à l'échelle de l'atome est la plupart du temps incompatible avec les puissances de calcul disponibles. C'est pourquoi des méthodes d'approximation sont indispensables pour pouvoir mettre en œuvre des simulations numériques à l'échelle de systèmes supramoléculaires réalistes. Dans le cadre de cette thèse, un modèle de dynamique moléculaire, dit « gros grain » a été mis au point à l'échelle du monomère de macromolécules pariétales. Les paramètres de ce modèle ont été calibrés à l'aide de simulations de dynamique moléculaire à l'échelle de l'atome. Ce modèle a fait l'objet de quatre applications : adsorption d'une chaine de xylane sur une surface de cellulose cristalline, arrachement d'une chaine de xylane adsorbée sur une surface de cellulose cristalline par une pointe AFM, adsorption d'une phase amorphe de xylane sur une surface de cellulose cristalline et adsorption d'une phase amorphe de xylane sur un monocristal de cellulose exposant trois surfaces différentes. Des effets de structuration au voisinage de la cellulose sont observés. / Nowadays, the understanding of plant cell walls' structure-properties relationship leans more and more on the use of molecular modeling approaches and of molecular dynamics in particular. To date, numerical weight of such an approach is usually out of the reach of available computing power if the atomic scale is used. As a consequence, building approximate methods is of crucial importance to perform numerical simulation of realistic supramolecular systems. Within the framework of this PhD, a “coarse grain” molecular dynamics model was built at plant cell wall macromolecule monomer's scale, it's parameters being fixed with the help of atom-scale molecular dynamics simulations. Then, several numerical studies were carried out: a single xylan chain was adsorbed on a crystalline cellulose surface, a single xylan chain was pulled from a crystalline cellulose surface with the help of the tip of an AFM cantilever, an amorphous xylan phase was adsorbed on a cellulose surface and an amorphous xylan phase was adsorbed on a cellulose crystal, which three surfaces were exposed. Local structuring effects were observed.

<p>In this thesis, we present novel computational methods and frameworks to address the challenges associated with the determination of free energy profiles for condensed-phase chemical reactions using combined quantum mechanical and molecular mechanical (QM/MM) approaches. We focus on overcoming issues related to force matching, molecular polarizability, and convergence of free energy profiles. First, we introduce a method called Reaction Path-Force Matching in Collective Variables (RP-FM-CV) that efficiently carries out ab initio QM/MM free energy simulations through mean force fitting. This method provides accurate and robust simulations of solution-phase chemical reactions by significantly reducing deviations on the collective variables forces, thereby bringing simulated free energy profiles closer to experimental and benchmark AI/MM results. Second, we explore the role of pairwise repulsive correcting potentials in generating converged free energy profiles for chemical reactions using QM/MM simulations. We develop a free energy correcting model that sheds light on the behavior of repulsive pairwise potentials with large force deviations in collective variables. Our findings contribute to a deeper understanding of force matching models, paving the way for more accurate predictions of free energy profiles in chemical reactions. Next, we address the underpolarization problem in semiempirical (SE) molecular orbital methods by introducing a hybrid framework called doubly polarized QM/MM (dp-QM/MM). This framework improves the response property of SE/MM methods through high-level molecular polarizability fitting using machine learning (ML)-derived corrective polarizabilities, referred to as chaperone polarizabilities. We demonstrate the effectiveness of the dp-QM/MM method in simulating the Menshutkin reaction in water, showing that ML chaperones significantly reduce the error in solute molecular polarizability, bringing simulated free energy profiles closer to experimental results. In summary, this thesis presents a series of novel methods and frameworks that improve the accuracy and reliability of free energy profile estimations in condensed-phase chemical reactions using QM/MM simulations. By addressing the challenges of force matching, molecular polarizability, and convergence, these advancements have the potential to impact various fields, including computational chemistry, materials science, and drug design.</p>

Indiana University-Purdue University Indianapolis (IUPUI) / In this thesis, we present novel computational methods and frameworks to address the challenges associated with the determination of free energy profiles for condensed-phase chemical reactions using combined quantum mechanical and molecular mechanical (QM/MM) approaches. We focus on overcoming issues related to force matching, molecular polarizability, and convergence of free energy profiles. First, we introduce a method called Reaction Path-Force Matching in Collective Variables (RP-FM-CV) that efficiently carries out ab initio QM/MM free energy simulations through mean force fitting. This method provides accurate and robust simulations of solution-phase chemical reactions by significantly reducing deviations on the collective variables forces, thereby bringing simulated free energy profiles closer to experimental and benchmark AI/MM results. Second, we explore the role of pairwise repulsive correcting potentials in generating converged free energy profiles for chemical reactions using QM/MM simulations. We develop a free energy correcting model that sheds light on the behavior of repulsive pairwise potentials with large force deviations in collective variables. Our findings contribute to a deeper understanding of force matching models, paving the way for more accurate predictions of free energy profiles in chemical reactions. Next, we address the underpolarization problem in semiempirical (SE) molecular orbital methods by introducing a hybrid framework called doubly polarized QM/MM (dp-QM/MM). This framework improves the response property of SE/MM methods through high-level molecular polarizability fitting using machine learning (ML)-derived corrective polarizabilities, referred to as chaperone polarizabilities. We demonstrate the effectiveness of the dp-QM/MM method in simulating the Menshutkin reaction in water, showing that ML chaperones significantly reduce the error in solute molecular polarizability, bringing simulated free energy profiles closer to experimental results. In summary, this thesis presents a series of novel methods and frameworks that improve the accuracy and reliability of free energy profile estimations in condensed-phase chemical reactions using QM/MM simulations. By addressing the challenges of force matching, molecular polarizability, and convergence, these advancements have the potential to impact various fields, including computational chemistry, materials science, and drug design.

No description available.