Many protein-protein interactions, especially those involved in eukaryotic signalling, are mediated by PDZ domains through the recognition of hydrophobic C-termini. The availability of experimental PDZ interaction data sets have led to the construction of computational methods to predict PDZ domain-peptide interactions. Such predictors are ideally suited to predict interactions in single organisms or for limited subsets of PDZ domains. As a result, the goal of my thesis has been to build general predictors that can be used to scan the proteomes of multiple organisms for ligands for almost all PDZ domains from select model organisms. A framework consisting of four steps: data collection, feature encoding, predictor training and evaluation was developed and applied for all predictors built in this thesis. The first predictor utilized PDZ domain-peptide sequence information from two interaction data sets obtained from high throughput protein microarray and phage display experiments in mouse and human, respectively. The second predictor used PDZ domain structure and peptide sequence information. I showed that these predictors are complementary to each other, are capable of predicting unseen interactions and can be used for the purposes of proteome scanning in human, worm and fly. As both positive and negative interactions are required for building a successful predictor, a major obstacle I addressed was the generation of artificial negative interactions for training. In particular, I used position weight matrices to generate such negatives for the positive only phage display data and used a semi-supervised learning approach to overcome the problem of over-prediction (i.e. prediction of too many positives). These predictors are available as a community web resource: http://webservice.baderlab.org/domains/POW. Finally, a Bayesian integration method combining information from different biological evidence sources was used to filter the human proteome scanning predictions from both predictors. This resulted in the construction of a comprehensive physiologically relevant high confidence PDZ mediated protein-protein interaction network in human.

Many protein-protein interactions, especially those involved in eukaryotic signalling, are mediated by PDZ domains through the recognition of hydrophobic C-termini. The availability of experimental PDZ interaction data sets have led to the construction of computational methods to predict PDZ domain-peptide interactions. Such predictors are ideally suited to predict interactions in single organisms or for limited subsets of PDZ domains. As a result, the goal of my thesis has been to build general predictors that can be used to scan the proteomes of multiple organisms for ligands for almost all PDZ domains from select model organisms. A framework consisting of four steps: data collection, feature encoding, predictor training and evaluation was developed and applied for all predictors built in this thesis. The first predictor utilized PDZ domain-peptide sequence information from two interaction data sets obtained from high throughput protein microarray and phage display experiments in mouse and human, respectively. The second predictor used PDZ domain structure and peptide sequence information. I showed that these predictors are complementary to each other, are capable of predicting unseen interactions and can be used for the purposes of proteome scanning in human, worm and fly. As both positive and negative interactions are required for building a successful predictor, a major obstacle I addressed was the generation of artificial negative interactions for training. In particular, I used position weight matrices to generate such negatives for the positive only phage display data and used a semi-supervised learning approach to overcome the problem of over-prediction (i.e. prediction of too many positives). These predictors are available as a community web resource: http://webservice.baderlab.org/domains/POW. Finally, a Bayesian integration method combining information from different biological evidence sources was used to filter the human proteome scanning predictions from both predictors. This resulted in the construction of a comprehensive physiologically relevant high confidence PDZ mediated protein-protein interaction network in human.

Computational drug repurposing aims at finding new medical uses for existing drugs. The identification of novel drug-target interactions (DTIs) can be a useful part of such a task. Computational determination of DTIs is a convenient strategy for systematic screening of a large number of drugs in the attempt to identify new DTIs at low cost and with reasonable accuracy. This necessitates development of accurate computational methods that can help focus on the follow-up experimental validation on a smaller number of highly likely targets for a drug. Although many methods have been proposed for computational DTI prediction, they suffer the high false positive prediction rate or they do not predict the effect that drugs exert on targets in DTIs. In this report, first, we present a comprehensive review of the recent progress in the field of DTI prediction from data-centric and algorithm-centric perspectives. The aim is to provide a comprehensive review of computational methods for identifying DTIs, which could help in constructing more reliable methods. Then, we present DDR, an efficient method to predict the existence of DTIs. DDR achieves significantly more accurate results compared to the other state-of-theart methods. As supported by independent evidences, we verified as correct 22 out of the top 25 DDR DTIs predictions. This validation proves the practical utility of DDR, suggesting that DDR can be used as an efficient method to identify 5 correct DTIs. Finally, we present DDR-FE method that predicts the effect types of a drug on its target. On different representative datasets, under various test setups, and using different performance measures, we show that DDR-FE achieves extremely good performance. Using blind test data, we verified as correct 2,300 out of 3,076 DTIs effects predicted by DDR-FE. This suggests that DDR-FE can be used as an efficient method to identify correct effects of a drug on its target.

Microarray technologies enable scientists to identify co-expressed genes at large scale. However, the gene expression analysis does not show functional relationships between co-expressed genes. There is a demand for effective approaches to analyse gene expression data to enable biological discoveries that can lead to identification of markers or therapeutic targets of many diseases. In cancer research, a number of gene expression screens have been carried out to identify genes differentially expressed in cancerous tissue such as Pancreatic Ductal Adenocarcinoma (PDAC). PDAC carries very poor prognosis, it eludes early detection and is characterised by its aggressiveness and resistance to currently available therapies. To identify molecular markers and suitable targets, there exist a research effort that maps differentially expressed genes to protein interactions to gain an understanding at systems level. Such interaction networks have a complex interconnected structure, whose the understanding of which is not a trivial task. Several formal approaches use simulation to support the investigation of such networks. These approaches suffer from the missing knowledge concerning biological systems. Reasoning in the other hand has the advantage of dealing with incomplete and partial information of the network knowledge. The initial approach adopted was to provide an algorithm that utilises a network-centric approach to pancreatic cancer, by re-constructing networks from known interactions and predicting novel protein interactions from structural templates. This method was applied to a data set of co-expressed PDAC genes. To this end, structural domains for the gene products are identified by using threading which is a 3D structure prediction technique. Next, the Protein Structure Interaction Database (SCOPPI), a database that classifies and annotates domain interactions derived from all known protein structures, is used to find templates of structurally interacting domains. Moreover, a network of related biological pathways for the PDAC data was constructed. In order to reason over molecular networks that are affected by dysregulation of gene expression, BioRevise was implemented. It is a belief revision system where the inhibition behaviour of reactions is modelled using extended logic programming. The system computes a minimal set of enzymes whose malfunction explains the abnormal expression levels of observed metabolites or enzymes. As a result of this research, two complementary approaches for the analysis of pancreatic cancer gene expression data are presented. Using the first approach, the pathways found to be largely affected in pancreatic cancer are signal transduction, actin cytoskeleton regulation, cell growth and cell communication. The analysis indicates that the alteration of the calcium pathway plays an important role in pancreas specific tumorigenesis. Furthermore, the structural prediction method reveals ~ 700 potential protein-protein interactions from the PDAC microarray data, among them, 81 novel interactions such as: serine/threonine kinase CDC2L1 interacting with cyclin-dependent kinase inhibitor CDKN3 and the tissue factor pathway inhibitor 2 (TFPI2) interacting with the transmembrane protease serine 4 (TMPRSS4). These resulting genes were further investigated and some were found to be potential therapeutic markers for PDAC. Since TMPRSS4 is involved in metastasis formation, it is hypothesised that the upregulation of TMPRSS4 and the downregulation of its predicted inhibitor TFPI2 plays an important role in this process. The predicted protein-protein network inspired the analysis of the data from two other perspectives. The resulting protein-protein interaction network highlighted the importance of the co-expression of KLK6 and KLK10 as prognostic factors for survival in PDAC as well as the construction of a PDAC specific apoptosis pathway to study different effects of multiple gene silencing in order to reactivate apoptosis in PDAC. Using the second approach, the behaviour of biological interaction networks using computational logic formalism was modelled, reasoning over the networks is enabled and the abnormal behaviour of its components is explained. The usability of the BioRevise system is demonstrated through two examples, a metabolic disorder disease and a deficiency in a pancreatic cancer associated pathway. The system successfully identified the inhibition of the enzyme glucose-6-phosphatase as responsible for the Glycogen storage disease type I, which according to literature is known to be the main reason for this disease. Furthermore, BioRevise was used to model reaction inhibition in the Glycolysis pathway which is known to be affected by Pancreatic cancer.

Un médicament, ou tout autre petite molécule biologique, agit sur l’organisme via des interactions chimiques qui se produisent avec d’autres macromolécules telles que les protéines qui régissent le fonctionnement des cellules. La détermination de l’ensemble des cibles, c’est à dire de l’ensemble des macromolécules susceptibles de lier une même molécule, est essentielle pour mieux comprendre les mécanismes moléculaires à l’origine des effets d’un médicament. Cette connaissance permettrait en effet de guider la conception d’un composé pour éviter au mieux les effets secondaires indésirables, ou au contraire découvrir de nouvelles applications à des molécules connues. Les avancées de la biologie structurale nous permettent maintenant d’avoir accès à un très grand nombre de structures tridimensionnelles de protéines impliquées dans ces interactions, ce qui motive l’utilisation d’outils in silico (informatique) pour complémenter ou guider les expériences in vitro ou in vivo plus longues et plus chères.La thèse s’inscrit dans le cadre d’une collaboration entre le laboratoire DAVID de l’Université de Versailles-Saint-Quentin, et l’entreprise Bionext SA qui propose une suite logicielle permettant de visualiser et d’étudier les interactions chimiques. Les travaux de recherches ont pour objectif de développer un algorithme permettant, à partir des données structurales des protéines, de déterminer des cibles potentielles pour un composé donné. L’approche choisie consiste à utiliser la connaissance d’une première interaction entre un composé et une protéine afin de rechercher par similarité d’autres protéines pour lesquelles on peut inférer la capacité à se lier avec le même composé. Il s’agit plus précisément de rechercher une similarité locale entre un motif donné, qui est la région permettant à la cible connue de lier le composé, et un ensemble de protéines candidates.Un algorithme a été développé, BioBind, qui utilise un modèle des surfaces des macromolécules issu de la théorie des formes alpha afin de modéliser la surface accessible ainsi qu’une topologie sur cette surface permettant la définition de régions en surface. Afin de traiter le problème de la recherche d’un motif en surface, une heuristique est utilisée consistant à définir des motifs réguliers qui sont une approximation de disques géodésiques et permettant un échantillonnage exhaustif à la surface des macromolécules. Ces régions circulaires sont alors étendues à l’ensemble du motif recherché afin de déterminer une mesure de similarité.Le problème de la prédiction de cibles est ramené à un problème de classification binaire, où il s’agit pour un ensemble de protéines données de déterminer lesquelles sont susceptibles d’interagir avec le composé considéré, par similarité avec la première cible connue. Cette formalisation permet d’étudier les performances de notre approche, ainsi que de la comparer avec d’autres approches sur différents jeux de données. Nous utilisons pour cela deux jeux de données issus de la littérature ainsi qu’un troisième développé spécifiquement pour cette problématique afin d’être plus représentatif des molécules pertinentes du point de vue pharmacologique, c’est-à-dire ayant des propriétés proches des médicaments. Notre approche se compare favorablement sur ces trois jeux de données par rapport à une autre approche de prédiction par similarité, et plus généralement notre analyse confirme que les approches par docking (amarrage) sont moins performantes que les approches par similarité pour le problème de la prédiction de cibles. / The action of a drug, or another small biomolecule, is induced by chemical interactions with other macromolecules such as proteins regulating the cell functions. The determination of the set of targets, the macromolecules that could bind the same small molecule, is essential in order to understand molecular mechanisms responsible for the effects of a drug. Indeed, this knowledge could help the drug design process so as to avoid side effects or to find new applications for known drugs. The advances of structural biology provides us with three-dimensional representations of many proteins involved in these interactions, motivating the use of in silico tools to complement or guide further in vitro or in vivo experiments which are both more expansive and time consuming.This research is conducted as part of a collaboration between the DAVID laboratory of the Versailles-Saint-Quentin University, and Bionext SA which offers a software suite to visualize and analyze chemical interactions between biological molecules. The objective is to design an algorithm to predict these interactions for a given compound, using the structures of potential targets. More precisely, starting from a known interaction between a drug and a protein, a new interaction can be inferred with another sufficiently similar protein. This approach consists in the search of a given pattern, the known binding site, across a collection of macromolecules.An algorithm was implemented, BioBind, which rely on a topological representation of the surface of the macromolecules based on the alpha shapes theory. Our surface representation allows to define a concept of region of any shape on the surface. In order to tackle the search of a given pattern region, a heuristic has been developed, consisting in the definition of regular region which is an approximation of a geodesic disk. This circular shape allows for an exhaustive sampling and fast comparison, and any circular region can then be extended to the actual pattern to provide a similarity evaluation with the query binding site.The target prediction problem is formalized as a binary classification problem, where a set of macromolecules is being separated between those predicted to interact and the others, based on their local similarity with the known target. With this point of view, classic metrics can be used to assess performance, and compare our approach with others. Three datasets were used, two of which were extracted from the literature and the other one was designed specifically for our problem emphasizing the pharmacological relevance of the chosen molecules. Our algorithm proves to be more efficient than another state-of-the-art similarity based approach, and our analysis confirms that docking software are not relevant for our target prediction problem when a first target is known, according to our metric.

Microarray technologies enable scientists to identify co-expressed genes at large scale. However, the gene expression analysis does not show functional relationships between co-expressed genes. There is a demand for effective approaches to analyse gene expression data to enable biological discoveries that can lead to identification of markers or therapeutic targets of many diseases. In cancer research, a number of gene expression screens have been carried out to identify genes differentially expressed in cancerous tissue such as Pancreatic Ductal Adenocarcinoma (PDAC). PDAC carries very poor prognosis, it eludes early detection and is characterised by its aggressiveness and resistance to currently available therapies. To identify molecular markers and suitable targets, there exist a research effort that maps differentially expressed genes to protein interactions to gain an understanding at systems level. Such interaction networks have a complex interconnected structure, whose the understanding of which is not a trivial task. Several formal approaches use simulation to support the investigation of such networks. These approaches suffer from the missing knowledge concerning biological systems. Reasoning in the other hand has the advantage of dealing with incomplete and partial information of the network knowledge. The initial approach adopted was to provide an algorithm that utilises a network-centric approach to pancreatic cancer, by re-constructing networks from known interactions and predicting novel protein interactions from structural templates. This method was applied to a data set of co-expressed PDAC genes. To this end, structural domains for the gene products are identified by using threading which is a 3D structure prediction technique. Next, the Protein Structure Interaction Database (SCOPPI), a database that classifies and annotates domain interactions derived from all known protein structures, is used to find templates of structurally interacting domains. Moreover, a network of related biological pathways for the PDAC data was constructed. In order to reason over molecular networks that are affected by dysregulation of gene expression, BioRevise was implemented. It is a belief revision system where the inhibition behaviour of reactions is modelled using extended logic programming. The system computes a minimal set of enzymes whose malfunction explains the abnormal expression levels of observed metabolites or enzymes. As a result of this research, two complementary approaches for the analysis of pancreatic cancer gene expression data are presented. Using the first approach, the pathways found to be largely affected in pancreatic cancer are signal transduction, actin cytoskeleton regulation, cell growth and cell communication. The analysis indicates that the alteration of the calcium pathway plays an important role in pancreas specific tumorigenesis. Furthermore, the structural prediction method reveals ~ 700 potential protein-protein interactions from the PDAC microarray data, among them, 81 novel interactions such as: serine/threonine kinase CDC2L1 interacting with cyclin-dependent kinase inhibitor CDKN3 and the tissue factor pathway inhibitor 2 (TFPI2) interacting with the transmembrane protease serine 4 (TMPRSS4). These resulting genes were further investigated and some were found to be potential therapeutic markers for PDAC. Since TMPRSS4 is involved in metastasis formation, it is hypothesised that the upregulation of TMPRSS4 and the downregulation of its predicted inhibitor TFPI2 plays an important role in this process. The predicted protein-protein network inspired the analysis of the data from two other perspectives. The resulting protein-protein interaction network highlighted the importance of the co-expression of KLK6 and KLK10 as prognostic factors for survival in PDAC as well as the construction of a PDAC specific apoptosis pathway to study different effects of multiple gene silencing in order to reactivate apoptosis in PDAC. Using the second approach, the behaviour of biological interaction networks using computational logic formalism was modelled, reasoning over the networks is enabled and the abnormal behaviour of its components is explained. The usability of the BioRevise system is demonstrated through two examples, a metabolic disorder disease and a deficiency in a pancreatic cancer associated pathway. The system successfully identified the inhibition of the enzyme glucose-6-phosphatase as responsible for the Glycogen storage disease type I, which according to literature is known to be the main reason for this disease. Furthermore, BioRevise was used to model reaction inhibition in the Glycolysis pathway which is known to be affected by Pancreatic cancer.

A biological function is rarely accomplished by a single gene. More often, proteins come together in complexes, and it is their collaboration within a complex that enables the associated biological function. However, the current map of the interactome is incomplete, meaning we have not observed all the interactions occurring in the cell yet. Gold standard experimental methods for the determination of all the protein-protein interactions (PPIs) in human interactome are time-consuming, expensive and may not even be feasible considering the vast number of protein pairs that need to be tested. For decades, scientists and engineers dedicated their efforts to forecasting protein interactions, predominantly relying on network topology methods. However, the emergence of AlphaFold2 intelligence has redefined the computational biology field by harnessing 3D molecular structural data to predict interacting protein in complexes, offering a promising alternative to traditional laboratory experiments. It is in this context that we introduce an innovative concept known as Network Shape Intelligence (NSI). It is the intelligence displayed by any topological network automata to perform valid connectivity predictions without training, but only processing the input knowledge associated to the local topological network organization. NSI transcends conventional link prediction methods by weaving together principles inspired by brain network science. It achieves this by minimizing external links within local communities, a strategy founded on local topology and plasticity principles initially developed for brain networks but subsequently extended to diverse complex networks. In addition to the incompleteness of the PPI network, the question of the reliability of the existing wealth of information through observed physical links also arises. Therefore, to evaluate the performance of a predictor we must make sure that the tested positive and negative interactions are reliable. We introduce the Bona Fide Evaluation Methodology (BFEM). The rigor of protein interaction predictions is ensured through a balanced classification scenario, meticulously constructed using the well-studied yeast protein interactome. Our methodology focuses on creating a golden standard set of true and false interactions, enhancing the reliability of our evaluations. We show that by using only local network information and without the need for training, these network automata designed for modelling and predicting network connectivity can outperform AlphaFold2 intelligence in vanilla protein interactions prediction. We find that the set of interactions mispredicted by AlphaFold2 predominantly consists of proteins whose amino acids exhibit higher probability of being associated with intrinsically disordered regions. Finally, we suggest that the future advancements in AlphaFold intelligence could integrate principles of NSI to further enhance the modelling and structural prediction of protein interactions.

Pattern discovery in protein structures is a fundamental task in computational biology, with important applications in protein structure prediction, profiling and alignment. We propose a novel approach for pattern discovery in protein structures using Particle Swarm-based flying windows over potentially promising regions of the search space. Using a heuristic search, based on Particle Swarm Optimization (PSO) is, however, easily trapped in local optima due to the sparse nature of the problem search space. Thus, we introduce a novel fitness-based stagnation detection technique that effectively and efficiently restarts the search process to escape potential local optima. The proposed fitness-based method significantly outperforms the commonly-used distance-based method when tested on eight classical and advanced (shifted/rotated) benchmark functions, as well as on two other applications for proteomic pattern matching and discovery. The main idea is to make use of the already-calculated fitness values of swarm particles, instead of their pairwise distance values, to predict an imminent stagnation situation. That is, the proposed fitness-based method does not require any computational overhead of repeatedly calculating pairwise distances between all particles at each iteration. Moreover, the fitness-based method is less dependent on the problem search space, compared with the distance-based method. The proposed pattern discovery algorithms are first applied to protein contact maps, which are the 2D compact representation of protein structures. Then, they are extended to work on actual protein 3D structures and interaction networks, offering a novel and low-cost approach to protein structure classification and interaction prediction. Concerning protein structure classification, the proposed PSO-based approach correctly distinguishes between the positive and negative examples in two protein datasets over 50 trials. As for protein interaction prediction, the proposed approach works effectively on complex, mostly sparse protein interaction networks, and predicts high-confidence protein-protein interactions — validated by more than one computational and experimental source — through knowledge transfer between topologically-similar interaction patterns of close proximity. Such encouraging results demonstrate that pattern discovery in protein structures and interaction networks are promising new applications of the fast-growing and far-reaching PSO algorithms, which is the main argument of this thesis. / Thesis (Ph.D, Computing) -- Queen's University, 2014-04-21 12:54:03.37

No description available.

Proteins are indispensable players in virtually all biological events. The functions of proteins are determined by their three dimensional (3D) structure and coordinated through intricate networks of protein-protein interactions (PPIs). Hence, a deep comprehension of such networks turns out to be crucial for understanding the cellular biology. Computational approaches have become critical tools for analysing PPI networks. In silico methods take advantage of the existing PPI knowledge to both predict new interactions and predict the function of proteins. Regarding the task of predicting PPIs, several methods have been already developed. However, recent findings demonstrate that such methods could take advantage of the knowledge on non-interacting protein pairs (NIPs). On the task of predicting the function of proteins,the Guilt-by-Association (GBA) principle can be exploited to extend the functional annotation of proteins over PPI networks. In this thesis, a new algorithm for PPI prediction and a protocol to complete cell signalling networks are presented. iLoops is a method that uses NIP data and structural information of proteins to predict the binding fate of protein pairs. A novel protocol for completing signalling networks –a task related to predicting the function of a protein, has also been developed. The protocol is based on the application of GBA principle in PPI networks. / Les proteïnes tenen un paper indispensable en virtualment qualsevol procés biològic. Les funcions de les proteïnes estan determinades per la seva estructura tridimensional (3D) i són coordinades per mitjà d’una complexa xarxa d’interaccions protiques (en anglès, protein-protein interactions, PPIs). Axí doncs, una comprensió en profunditat d’aquestes xarxes és fonamental per entendre la biologia cel•lular. Per a l’anàlisi de les xarxes d’interacció de proteïnes, l’ús de tècniques computacionals ha esdevingut fonamental als darrers temps. Els mètodes in silico aprofiten el coneixement actual sobre les interaccions proteiques per fer prediccions de noves interaccions o de les funcions de les proteïnes. Actualment existeixen diferents mètodes per a la predicció de noves interaccions de proteines. De tota manera, resultats recents demostren que aquests mètodes poden beneficiar-se del coneixement sobre parelles de proteïnes no interaccionants (en anglès, non-interacting pairs, NIPs). Per a la tasca de predir la funció de les proteïnes, el principi de “culpable per associació” (en anglès, guilt by association, GBA) és usat per extendre l’anotació de proteïnes de funció coneguda a través de xarxes d’interacció de proteïnes. En aquesta tesi es presenta un nou mètode pre a la predicció d’interaccions proteiques i un nou protocol basat per a completar xarxes de senyalització cel•lular. iLoops és un mètode que utilitza dades de parells no interaccionants i coneixement de l’estructura 3D de les proteïnes per a predir interaccions de proteïnes. També s’ha desenvolupat un nou protocol per a completar xarxes de senyalització cel•lular, una tasca relacionada amb la predicció de les funcions de les proteïnes. Aquest protocol es basa en aplicar el principi GBA a xarxes d’interaccions proteiques.