Search

National audience; Les fonctions biologiques dans la cellule mettent en jeu des interactions 3D entre protéines et ARN. Les avancées des techniques exérimentales restent insuffisantes pour de nombreuse applications. Il faut alors pouvoir prédire in silico les interactions protéine-ARN. Dans ce contexte, nos travaux sont focalisés sur la construction de fonctions de score permettant d'ordonner les solutions générées par le programme d'amarrage protéine-ARN RosettaDock. La méthodologie d'évaluation utilisée par RosettaDock impose de trouver une fonction de score s'exprimant comme une combinaison linéaire de mesures physicochimiques. Avec une approche d'apprentissage supervisé par algorithme génétique, nous avons appris différentes fonctions de score en imposant des contraintes sur la nature des poids recherchés. Les résultats obtenus montrent l'importance de la signification des poids à apprendre et de l'espace de recherche associé.

The biological function of macromolecules, such as proteins and nucleic acids, relies heavily on their interactions with their partners. The prediction of how molecules interact and how they can create large assemblies acting as nanomachines is essential for our understanding of biology but also for therapeutics and nanotechnology design.Blind challenges in biology, such as the CAPRI worldwide experiment for docking, have shown that in silico studies and simulations, mainly using physics-based potentials and techniques, could give structural insights in atomic detail. They might however be of very limited accuracy, particularly in predicting the native molecular structure of proteins, RNAs and complexes.Resorting to simple geometric coarse-grained modelling and machine learning strategies, such as genetic algorithms and support vector machines, we have shown that scoring the putative complex structures can be very much improved to reach the accuracy needed for experiment design and analysis, at least in a semi-rigid body context. From that proof of concept studies, most of the prediction strategies for docking now use machine learning for scoring optimization.Being able to predict the structure and the way molecular partners deform upon binding is also key to obtain better predictions, in particular for non-coding RNAs that are essential to target oncogenes. Our efforts in RNA structure prediction techniques have shown that data based parameterization of energy functions and statistical techniques largely improve the accuracy of structure prediction.Reaching the large assemblies stage also requires to be able to assess the dynamics of molecules from partial experimental data. We developed an efficient sampling technique based on inverse kinematics that does not rely on constraint counting, and implicitly calculates the rigidity of the molecule. Paired with experimental data, it offers an integrative view of the dynamics of non-coding RNAs for biological processes. Combined with clustering techniques, it allows for efficient and flexible cross-docking analysis for protein-RNA complexes.

The biological function of macromolecules, such as proteins and nucleic acids, relies heavily on their interactions with their partners. The prediction of how molecules interact and how they can create large assemblies acting as nanomachines is essential for our understanding of biology but also for therapeutics and nanotechnology design.Blind challenges in biology, such as the CAPRI worldwide experiment for docking, have shown that in silico studies and simulations, mainly using physics-based potentials and techniques, could give structural insights in atomic detail. They might however be of very limited accuracy, particularly in predicting the native molecular structure of proteins, RNAs and complexes.Resorting to simple geometric coarse-grained modelling and machine learning strategies, such as genetic algorithms and support vector machines, we have shown that scoring the putative complex structures can be very much improved to reach the accuracy needed for experiment design and analysis, at least in a semi-rigid body context. From that proof of concept studies, most of the prediction strategies for docking now use machine learning for scoring optimization.Being able to predict the structure and the way molecular partners deform upon binding is also key to obtain better predictions, in particular for non-coding RNAs that are essential to target oncogenes. Our efforts in RNA structure prediction techniques have shown that data based parameterization of energy functions and statistical techniques largely improve the accuracy of structure prediction.Reaching the large assemblies stage also requires to be able to assess the dynamics of molecules from partial experimental data. We developed an efficient sampling technique based on inverse kinematics that does not rely on constraint counting, and implicitly calculates the rigidity of the molecule. Paired with experimental data, it offers an integrative view of the dynamics of non-coding RNAs for biological processes. Combined with clustering techniques, it allows for efficient and flexible cross-docking analysis for protein-RNA complexes.

International audience; The following sections are included: Background, Session Summary and Acknowledgements.

The CAPRI and CASP prediction experiments have demonstrated the power of community wide tests of methodology in assessing the current state of the art and spurring progress in the very challenging areas of protein docking and structure prediction. We sought to bring the power of community wide experiments to bear on a very challenging protein design problem that provides a complementary but equally fundamental test of current understanding of protein-binding thermodynamics. We have generated a number of designed protein-protein interfaces with very favorable computed binding energies but which do not appear to be formed in experiments, suggesting there may be important physical chemistry missing in the energy calculations. 28 research groups took up the challenge of determining what is missing: we provided structures of 87 designed complexes and 120 naturally occurring complexes and asked participants to identify energetic contributions and/or structural features that distinguish between the two sets. The community found that electrostatics and solvation terms partially distinguish the designs from the natural complexes, largely due to the non-polar character of the designed interactions. Beyond this polarity difference, the community found that the designed binding surfaces were on average structurally less embedded in the designed monomers, suggesting that backbone conformational rigidity at the designed surface is important for realization of the designed function. These results can be used to improve computational design strategies, but there is still much to be learned; for example, one designed complex, which does form in experiments, was classified by all metrics as a non-binder.

International audience; Knowledge-based protein folding potentials have proven successful in the recent years. Based on statistics of observed interatomic distances, they generally encode pairwise contact information. In this study we present a method that derives multi-body contact potentials from measurements of surface areas using coarse-grained protein models. The measurements are made using a newly implemented geometric construction: the arrangement of circles on a sphere. This construction allows the definition of residue covering areas which are used as parameters to build functions able to distinguish native structures from decoys. These functions, encoding up to 5-body contacts are evaluated on a reference set of 66 structures and its 45000 decoys, and also on the often used lattice ssfit set from the decoys'R us database. We show that the most relevant information for discrimination resides in 2- and 3-body contacts. The potentials we have obtained can be used for evaluation of putative structural models; they could also lead to different types of structure refinement techniques that use multi-body interactions.

The function of a protein is often subordinated to its interaction with one or many partners. Yet, the tridimensional structure study of this complexes, that can't be done experimentally, would permit the understanding of many cellular processes. This work contains two parts. The first part concerns the setting up of a scoring function for protein-protein docking and the second part concerns the crystallographic structure study of a tetrameric protein : the Paramecium Bursaria Chlorella Virus thymidylate synthase X, a potential antibacterial target. Docking of protein-protein complexes consists in two successive steps : first a large number of putative conformations are generated, then a scoring function is applied to rank them. This scoring function has to take into account both geometric complementarity of the two molecules and physico-chemical properties of surfaces in interaction. We addressed the second step of this problem through the development of a quick and reliable scoring function. This was done using Voronoi tessellation of the tridimensional structure of the proteins. Voronoi or Laguerre tessellations were shown to be good mathematical models of protein structure. In particular, this formalization leads to a good description of structural properties of the residues. This modeling illustrates the packing of the residues at the interface between two proteins. Thus, it is possible to measure a set of parameters, on protein-protein complexes whose structure is known, and on decoys. These parameters are frequencies of residues and pair frequencies of the residues at the interface, volumes of Voronoi cells, distances between residues at the interface, interface area and number of residues at the interface. They were used as input in statistical machine learning procedures (logistic learning, support vector machines (SVM) and genetic algorithms). These led to efficient scoring functions, able to separate native structures from decoys. In the second part, I describe the experimental determination of thymidylate synthase X tridimensionnal structure, an interesting antibacterial target. Thymidylate synthase X is a flavoprotein discovered recently. It plays a key role in the synthesis of dTMP in most of the prokaryotic organisms, but does not exist in superior eukaryotic organisms. This protein catalyses the methyl transfer from tetrahydrofolate to dUMP using FAD as a cofactor and NADPH as substrate. The tridimensional structure of ThyX homotetramer with its cofactor, FAD, was solved at 2.4Å by molecular replacement. As shown in the Thermotoga maritima and Mycobacterium tuberculosis ThyX structures, the monomer contains a core of β sheets and two α helices at its extremity. The active site is at the interface between three monomers, the isoalloxazine part of FAD being accessible to the solvent and close to a long flexible loop. FAD binding in this structure is a little different from those already observed, especially its the adenine part. This structure, in association with directed mutagenesis experiments made by our collabora- tors, revealed residues playing a key role during the catalysis.; La fonction d'une protéine est souvent subordonnée à l'interaction avec un certain nombre de partenaires. L'étude de la structure tridimensionnelle de ces complexes, qui ne peut souvent se faire expérimentalement, permettrait la compréhension de nombreux processus cellulaires. Le travail présenté ici se compose de deux parties. La première traite de la mise en place d'une fonction de score pour l'amarrage protéine-protéine et la deuxième de l'étude cristallographique d'une protéine tétramérique qui est une cible antibiotique potentielle : la thymidylate synthase X de Paramecium bursaria Chlorella virus. La modélisation des complexes protéine-protéine ou docking comporte deux étapes successives : d'abord, un grand nombre de conformations sont générées, puis une fonction de score est utilisée pour les classer. Cette fonction de score doit prendre en compte à la fois la complémentarité géométrique des deux molécules et les propriétés physico-chimiques des surfaces en interaction. Nous nous sommes intéressés à la seconde étape à travers le développement d'une fonction de score rapide et fiable. Ceci est possible grâce à la tessellation de Voronoï de la structure tridimensionnelle des protéines. En effet, les tessellations de Voronoï ou de Laguerre se sont avérées être de bons modèles mathématiques de la structure des protéines. En particulier, cette formalisation permet de faire une bonne description de l'empilement et des propriétés structurales des résidus. Cette modélisation rend compte l'empilement des résidus à l'interface entre deux protéines. Ainsi, il est possible de mesurer un ensemble de paramètres sur des complexes protéine-protéine dont la structure est connue expérimentalement et sur des complexes leurres générés artificiel- lement. Ces paramètres, sont la fréquence d'apparition des résidus ou des paires de résidus, les volumes des cellules de Voronoï, les distances entre les résidus en contact à l'interface, la surface de l'interface et le nombre de résidus à l'interface. Ils ont été utilisés en entrée de procédures d'apprentissage statistique. Grâce à ces procédures (apprentissage logistique, séparateurs à vaste marge (SVM) et algorithmes génétiques), on peut obtenir des fonctions de score efficaces, ca- pables de séparer les leurres des structures réelles. Dans un deuxième temps, j'ai déterminé expérimentalement la structure de la thymidylate synthase X, cible antibiotique de choix. La thymidylate synthase X est une flavoprotéine qui a été découverte récemment. Elle intervient dans la synthèse du dTMP chez la plupart des procaryotes mais n'existe pas chez les eucaryotes supérieurs. Cette protéine catalyse le transfert de methyle du tétrahydrofolate vers le dUMP grâce à son cofacteur le FAD et au NADPH qui intervient comme substrat. La structure tridimensionnelle de l'homotétramère de la thymidylate synthase X en présence de son cofacteur, le FAD, a été résolue à 2.4 Å par remplacement moléculaire. Comme pour les structures de thymidylate synthase X de Thermotoga maritima et de Mycobacterium tuberculosis précédemment résolues, le monomère se compose d'un coeur de feuillets β et de deux hélices α à son extrémité. Le site actif se trouve à l'interface de trois monomères, la partie isoalloxazine du FAD étant accessible au solvant et proche d'une longue boucle flexible. La fixation du FAD dans cette structure est légèrement différente de celles déjà observées par la conformation de la partie adénine. Cette structure, associée aux études de mutagénèse dirigée de nos collaborateurs, a permis de mettre évidence des résidus jouant un rôle majeur lors de la catalyse.

International audience; Dans cet article, nous décrivons l'interaction protéine-protéine dans le cadre du modèle de l'énergie stochastique (Random Energy Model) de la physique statistique. Nous simulons en amarrant les deux protéines partenaires, le processus d'association de deux protéines formant un complexe. Nous obtenons les spectres d'énergie d'un jeu de complexes protéine-protéine de structure tridimensionnelle connue en effectuant l'amarrage dans des orientations aléatoires et en calculant un score pour les complexes modèles ainsi générés. Nous utilisons une représentation très simplifiée de la structure où chaque acide aminé est remplacé par sa cellule de Voronoï, et nous appliquons le programme d'apprentissage par algorithme génétique ROGER à un ensemble de paramètres mesurés sur cette représentation, afin de dériver une fonction de score appropriée. En considérant les scores obtenus comme des énergies d'interaction, nous obtenons le spectre d'énergie de chaque complexe. Il s'approche d'une distribution gaussienne qui nous permet de calculer des paramètres physiques comme la température de transition vitreuse et la température de transition spécifique du système.

International audience; No abstract received.