Enzymes are ubiquitous in biological systems. They catalyze chemical reactions and are involved in many biochemical processes. The enzyme of interest is Pseudomonas aeruginosa D-arginine dehydrogenase (PaDADH). This flavin-dependent enzyme is composed of approximately 375 amino acid residues and has a broad substrate specificity with D-amino acids. A water recognition motif, observed in roughly 1200 non-redundant protein data bank (PDB) structures, was revealed to be embedded near the active site of PaDADH. This motif coincides with the conformational changes of the enzyme’s gated mechanism. Molecular dynamics simulations were carried out to study the gated properties and structural characteristics of PaDADH in solution. Single amino acid mutations were undertaken to further understand the dynamics of the gated mechanism of this enzyme. In addition, pKa,shift analyses were evaluated to probe for the basic catalytic amino acid residue that is suggested to trigger the catalytic mechanism of PaDADH.

The Human Genome Project, initiated in 1990, creates an enormous amount of excitement in human genetics—a field of study that seeks answers to the understanding of human evolution, diseases and development, gene therapy, and preventive medicine. The first completion of a human genome in 2003 and the breakthroughs of sequencing technologies in the past few years deliver the promised benefits of genome studies, especially in the roles of genomic variability and human health. However, intensive resource requirements and the associated costs make it infeasible to experimentally verify the effect of every genetic variation. At this stage of genome studies, in silico predictions play an important role in identifying putative functional variants. The most common practice for genome variant evaluation is based on the evolutionary conservation at the mutation site. Nonetheless, sequence conservation is not the absolute predictor for deleteriousness since phylogenetic diversity of aligned sequences used to construct the prediction algorithm has substantial effects on the analysis. This dissertation aims at overcoming the weaknesses of the conservation-based assumption for predicting the variant effects. The dissertation describes three different integrative computational approaches to identify a subset of high-priority amino acid mutations, derived from human genome data. The methods investigate variant-function relationships in three aspects of genome studies—personal genomics, genomics of epilepsy disorders, and genomics of variable drug responses. For genetic variants found in genomes of healthy individuals, an eight-level variant classification scheme is implemented to rank variants that are important towards individualized health profiles. For candidate genetic variants of epilepsy disorders, a novel 3-dimensional structure-based assessment protocol for amino acid mutations is established to improve discrimination between neutral and causal variants at less conserved sites, and to facilitate variant prioritization for experimental validations. For genomic variants that may affect inter-individual variability in drug responses, an explicit structure-based predictor for structural disturbances is developed to efficiently evaluate unknown variants in pharmacogenes. Overall, the three integrative approaches provide an opportunity for examining the effects of genomic variants from multiple perspectives of genome studies. They also introduce an efficient way to catalog amino acid variants on a large scale genome data.

This thesis is focused on predicting the impact of amino acid substitution on protein stability. The main goal is to create a consensual predictor that uses the outputs of chosen existing tools in order to improve accuracy of prediction. The optimal consensus of theese tools was designed using evolution strategies in three variants: 1/5 success rule, self-adaptation variant and the CMA-ES method. Then, the quality of calculated weight vectors was tested on the independent dataset. Although the highest prediction performance was attained by self-adaptation method, the differences between all three variants were not significant. Compared to the individual tools, the predictions provided by consensual methods were generally more accurate - the self-adaptation variant imporved the Pearson's corelation coeficient of the predictions by 0,057 on the training dataset. On the testing dataset, the improvement of designed method was smaller (0,040). Relatively low improvement of prediction performance (both on the training and the testing dataset) were caused by the fact, that for some records of testing dataset, some individual tools vere not able to provide their results. When omitting these records, consensual method improved the Pearson's corelations coeficient by 0,118.