<p>Proteins play a crucial roll in all biological processes. The wide range of protein functions is made possible through the many different conformations that the protein chain can adopt. The structure of a protein is extremely important for its function, but to determine the structure of protein experimentally is both difficult and time consuming. In fact with the current methods it is not possible to study all the billions of proteins in the world by experiments. Hence, for the vast majority of proteins the only way to get structural information is through the use of a method that predicts the structure of a protein based on the amino acid sequence.</p><p>This thesis focuses on improving the current protein structure prediction methods by combining different prediction approaches together with machine-learning techniques. This work has resulted in some of the best automatic servers in world – Pcons and Pmodeller. As a part of the improvement of our automatic servers, I have also developed one of the best methods for predicting the quality of a protein model – ProQ. In addition, I have also developed methods to predict the local quality of a protein, based on the structure – ProQres and based on evolutionary information – ProQprof. Finally, I have also performed the first large-scale benchmark of publicly available homology modeling programs.</p>

Proteins play a crucial roll in all biological processes. The wide range of protein functions is made possible through the many different conformations that the protein chain can adopt. The structure of a protein is extremely important for its function, but to determine the structure of protein experimentally is both difficult and time consuming. In fact with the current methods it is not possible to study all the billions of proteins in the world by experiments. Hence, for the vast majority of proteins the only way to get structural information is through the use of a method that predicts the structure of a protein based on the amino acid sequence. This thesis focuses on improving the current protein structure prediction methods by combining different prediction approaches together with machine-learning techniques. This work has resulted in some of the best automatic servers in world – Pcons and Pmodeller. As a part of the improvement of our automatic servers, I have also developed one of the best methods for predicting the quality of a protein model – ProQ. In addition, I have also developed methods to predict the local quality of a protein, based on the structure – ProQres and based on evolutionary information – ProQprof. Finally, I have also performed the first large-scale benchmark of publicly available homology modeling programs.

Protein structure prediction is concerned with the prediction of a protein's three dimensional structure from its amino acid sequence. Such predictions are commonly performed by searching the possible structures and evaluating each structure by using some scoring function. If it is assumed that the target protein structure resembles the structure of a known protein, the search space can be significantly reduced. Such an approach is referred to as comparative structure prediction. When such an assumption is not made, the approach is known as ab initio structure prediction. There are several difficulties in devising efficient searches or in computing the scoring function. Many of these problems have ready solutions from known mathematical methods. However, the problems that are yet unsolved have hindered structure prediction methods from more ideal predictions. The objective of this study is to present a complete framework for ab initio protein structure prediction. To achieve this, a new search strategy is proposed, and better techniques are devised for computing the known scoring functions. Some of the remaining problems in protein structure prediction are revisited. Several of them are shown to be intractable. In many of these cases, approximation methods are suggested as alternative solutions. The primary issues addressed in this thesis are concerned with local structures prediction, structure assembly or sampling, side chain packing, model comparison, and structural alignment. For brevity, we do not elaborate on these problems here; a concise introduction is given in the first section of this thesis. Results from these studies prompted the development of several programs, forming a utility suite for ab initio protein structure prediction. Due to the general usefulness of these programs, some of them are released with open source licenses to benefit the community.

Protein structure prediction is concerned with the prediction of a protein's three dimensional structure from its amino acid sequence. Such predictions are commonly performed by searching the possible structures and evaluating each structure by using some scoring function. If it is assumed that the target protein structure resembles the structure of a known protein, the search space can be significantly reduced. Such an approach is referred to as comparative structure prediction. When such an assumption is not made, the approach is known as ab initio structure prediction. There are several difficulties in devising efficient searches or in computing the scoring function. Many of these problems have ready solutions from known mathematical methods. However, the problems that are yet unsolved have hindered structure prediction methods from more ideal predictions. The objective of this study is to present a complete framework for ab initio protein structure prediction. To achieve this, a new search strategy is proposed, and better techniques are devised for computing the known scoring functions. Some of the remaining problems in protein structure prediction are revisited. Several of them are shown to be intractable. In many of these cases, approximation methods are suggested as alternative solutions. The primary issues addressed in this thesis are concerned with local structures prediction, structure assembly or sampling, side chain packing, model comparison, and structural alignment. For brevity, we do not elaborate on these problems here; a concise introduction is given in the first section of this thesis. Results from these studies prompted the development of several programs, forming a utility suite for ab initio protein structure prediction. Due to the general usefulness of these programs, some of them are released with open source licenses to benefit the community.

No description available.

The most significant impediment for protein structure prediction is the inadequacy of conformation space search. Conformation space is too large and the energy landscape too rugged for existing search methods to consistently find near-optimal minima. Conformation space search methods thus have to focus exploration on a small fraction of the search space. The ability to choose appropriate regions, i.e. regions that are highly likely to contain the native state, critically impacts the effectiveness of search. To make the choice of where to explore requires information, with higher quality information resulting in better choices. Most current search methods are designed to work in as many domains as possible, which leads to less accurate information because of the need for generality. However, most domains provide unique, and accurate information. To best utilize domain specific information search needs to be customized for each domain. The first contribution of this thesis customizes search for protein structure prediction, resulting in significantly more accurate protein structure predictions. Unless information is perfect, mistakes will be made, and search will focus on regions that do not contain the native state. How search recovers from mistakes is critical to its effectiveness. To recover from mistakes, this thesis introduces the concept of adaptive balancing of exploitation with exploration. Adaptive balancing of exploitation with exploration allows search to use information only to the extent to which it guides exploration toward the native state. Existing methods of protein structure prediction rely on information from known proteins. Currently, this information is from either full-length proteins that share similar sequences, and hence have similar structures (homologs), or from short protein fragments. Homologs and fragments represent two extremes on the spectrum of information from known proteins. Significant additional information can be found between these extremes. However, current protein structure prediction methods are unable to use information between fragments and homologs because it is difficult to identify the correct information from the enormous amount of incorrect information. This thesis makes it possible to use information between homologs and fragments by adaptively balancing exploitation with exploration in response to an estimate of template protein quality. My results indicate that integrating the information between homologs and fragments significantly improves protein structure prediction accuracy, resulting in several proteins predicted with <1>°A RMSD resolution.

P2X receptors (P2XRs) are trimeric cation channels activated by extracellular ATP. Human P2XRs (P2X1-7) are expressed in nearly all mammalian tissues, and they are an important drug target because of their involvement in inflammation and neuropathic pain. The aim of this thesis is to address the following questions. P2XR crystal structures have revealed an unusual U-shape conformation for bound ATP; how does the U-shape conformation of ATP and its derivatives affect channel activation? Where and how do the selective, non-competitive inhibitors AZ10606120 and A438079 bind to P2X7R? What is the structure of the hP2X1R intracellular domain in the closed state? Molecular modelling and bioinformatics were used to answer these questions, hypotheses resulting from this work were tested in collaboration with Prof. Evans. Investigating the binding modes of ATP and its deoxy forms in hP2X1R showed that the ribose 2′-hydroxyl group is stabilising the U-shape conformation by a hydrogen bond to the γ-phosphate. The reduced ability of 2′-deoxy ATP to adopt the U-shape conformation could explain its weak agonist action in contrast to full agonists ATP and 3′-deoxy ATP. Ligand docking of AZ10606120 and A438079 into the hP2X7R predicted an allosteric binding site, this site has meanwhile been confirmed by P2X7R/antagonist X-ray structures. MD simulations suggested that unique P2X7R regions (residues 73-79 and T90/T94) contribute to an increase of the allosteric pocket volume compared to the hP2X1R. This difference in size might be the key for selectivity. The hP2X1R intracellular domain in the closed state was modelled ab initio, and interpreted in context of chemical cross-links (collaboration with Prof. Evans). This suggests a symmetrical arrangement of two short b-antiparallel strands within the Nterminal region and short a-helix in the C-terminal region and additional asymmetrical states.

Fragment assembly methods represent the state of the art in computational protein structure prediction. However, one limitation of these methods, particularly for larger protein structures, is inadequate conformational sampling. This thesis describes studies aimed at uncovering potential causes of ineffective sampling, and the development of methods to try and address these problems. To identify behaviours that might lead to poor conformational sampling, we developed measures to study fragment-based sampling trajectories. Applying these measures to the Rosetta Abinitio and EdaFold methods showed similarities and differences in the ways that these methods make predictions, and pointed to common limitations. In both protocols, structural features such as alpha-helices were more frequently altered during the search, as compared with regions such as loops. Analyses of the fragment libraries used by these methods showed that fragments covering loop regions were less likely to possess native-like structural features, and this likely exacerbated the problems of inadequate sampling in these regions. Inadequate loop sampling leads to poor fold-level exploration within individual runs of methods such as Rosetta, and this necessitates the use of many independent runs. Guided by these findings, we developed new heuristic-based search algorithms. These algorithms were designed to facilitate the exploration of multiple energy basins within runs. Over many runs, the enhanced exploration in our protocols produced decoy sets with larger fractions of native-like solutions as compared to runs of Rosetta. Experiments with different fragment sets indicated that our methods could better translate increased fragment set quality into improvements in predictive accuracy distributions. These improvements depend most strongly on the ability of search algorithms to reliably generate native-like structures using a fragment set. In contrast, inadequate retention of native-like decoys when associated with unfavourable score values appears to be less of an issue. This thesis shows that targeted developments in conformational sampling strategies can improve the accuracy and reliability of predictions. With effective conformational sampling methods, developments in methods for fragment set construction and other areas may more reliably enhance predictive ability.

Of the 20 standard amino acids, cysteines are the only amino acids that have a reactive sulphur atom, thus enabling two cysteines to form strong covalent bonds known as disulfide bonds. Even though almost all proteins have cysteines, not all of them have disulfide bonds. Disulfide bonds provide structural stability to proteins and hence are an important constraint in determining the structure of a protein. As a result, disulfide bonds are used to study various protein properties, one of them being protein folding. Protein structure prediction is the problem of predicting the three-dimensional structure of a protein from its one-dimensional amino acid sequence. Ab initio methods are a group of methods that attempt to solve this problem from first principles, using only basic physico-chemical properties of proteins. These methods use structure libraries of short amino acid fragments in the process of predicting the structure of a protein. The protein structures from which these structure libraries are created are not classified in any other way apart from being non-redundant. In this thesis, we investigate the structural dissimilarities of short amino acid fragments when occurring in proteins with disulfide bonds and when occurring in those proteins without disulfide bonds. We are interested in this because, as mentioned earlier, the protein structures from which the structure libraries of ab initio methods are created, are not classified in any form. This means that any significant structural difference in amino acids and short fragments when occurring in proteins with and without disulfide bonds would remain unnoticed as these structure libraries have both fragments from proteins with disulfide bonds and without disulfide bonds together. Our investigation of structural dissimilarities of amino acids and short fragments is done in four phases. In phase one, by statistically analysing the phi and psi backbone dihedral angle distributions we show that these fragments have significantly different structures in terms of dihedral angles when occurring in proteins with and without disulfide bonds. In phase two, using directional statistics we investigate how structurally different are the 20 different amino acids and the short fragments when occurring in proteins with and without disulfide bonds. In phase three of our work, we investigate the differences in secondary structure preference of the 20 amino acids in proteins with and without disulfide bonds. In phase four, we further investigate and show that there are significant differences within the same secondary structure region of amino acids when they occur in proteins with and without disulfide bonds. Finally, we present the design and implementation details of a dihedral angle and secondary structure database of short amino acid fragments (DASSD) that is publicly available. Thus, in this thesis we show previously unknown significant structure differences in terms of backbone dihedral angles and secondary structures in amino acids and short fragments when they occur in proteins with and without disulfide bonds.

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.