Predicting protein structure using its primary sequence has always been a challenging topic in biochemistry. Although it seems as simple as finding the minimal energy conformation, it has been quite difficult to provide an accurate yet reliable solution for the problem. On the one hand, the lack of understanding of the hydrophobic effect as well as the relationship between different stabilizing forces, such as hydrophobic interaction, hydrogen bonding and electronic static interaction prevent the scientist from developing potential functions to estimate free energy. On the other hand, structure databases are limited with redundant structures, which represent a noncontinuous, sparsely-sampled conformational space, and preventing the development of a method suitable for high-resolution, high-accuracy structure prediction that can be applied for functional annotation of an unknown protein sequence. Thus, in this study, we use molecular dynamics simulation as a tool to sample conformational space. Structures were generated with physically realistic conformations that represented the properties of ensembles of native structures. First, we focused our study on the relationship among different factors that stabilize protein structure. Using a wellcharacterized mutation system of the B-hairpin, a fundamental building block of protein, we were able to identify the effect of terminal ion-pairs (salt-bridges) on the stability of the beta-hairpin, and its relationship with hydrophobic interactions and hydrogen bonds. In the same study, we also correlated our theoretical simulations qualitatively with experimental results. Such analysis provides us a better understanding of beta-hairpin stability and helps us to improve the protein engineering method to design more stable hairpins. Second, with large-scale simulations of different representative protein folds, we were able to conduct a fine-grained analysis by sampling the continuous conformational space to characterize the relationship among backbone conformation, side-chain conformation and side-chain packing. Such information is valuable for improving high-resolution structure prediction. Last, with this information, we developed a new prediction algorithm using packing information derived from the conserved relative packing groups. Based on its performance in CASP7, we were able to draw the conclusion that our simulated dataset as well as our packing-oriented prediction method are useful for template based structure prediction.

Im Rahmen dieser Arbeit wird eine Vielzahl von Daten zur Struktur und Funktion von Proteinen gesammelt. Anschließend wird in strukturellen Daten die atomare Packungsdichte untersucht. Untersuchungen an Strukturen benötigen oftmals maßgeschneiderte Datensätze von Proteinen. Kriterien für die Auswahl einzelner Proteine sind z.B. Eigenschaften der Sequenzen, die Faltung oder die Auflösung einer Struktur. Solche Datensätze mit den im Netz verfügbaren Mitteln herzustellen ist mühselig, da die notwendigen Daten über viele Datenbanken verteilt liegen. Um diese Aufgabe zu vereinfachen, wurde Columba, eine integrierte Datenbank zur Annotation von Proteinstrukturen, geschaffen. Columba integriert insgesamt sechzehn Datenbanken, darunter u.a. die PDB, KEGG, Swiss-Prot, CATH, SCOP, die Gene Ontology und ENZYME. Von den in Columba enthaltenen Strukturen der PDB sind zwei Drittel durch viele andere Datenbanken annotiert. Zum verbliebenen Drittel gibt es nur wenige zusätzliche Angaben, teils da die entsprechenden Strukturen erst seit kurzem in der PDB sind, teils da es gar keine richtigen Proteine sind. Die Datenbank kann über eine Web-Oberfläche unter www.columba-db.de spezifisch für einzelne Quelldatenbanken durchsucht werden. Ein Benutzer kann sich auf diese Weise schnell einen Datensatz von Strukturen aus der PDB zusammenstellen, welche den gewählten Anforderungen entsprechen. Es wurden Regeln aufgestellt, mit denen Datensätze effizient erstellt werden können. Diese Regeln wurden angewandt, um Datensätze zur Analyse der Packungsdichte von Proteinen zu erstellen. Die Packungsanalyse quantifiziert den Raum zwischen Atomen, und kann Regionen finden, in welchen eine hohe lokale Beweglichkeit vorliegt oder welche Fehler in der Struktur beinhalten. In einem Referenzdatensatz wurde so eine große Zahl von atomgroßen Höhlungen dicht unterhalb der Proteinoberfläche gefunden. In Transmembrandomänen treten diese Höhlungen besonders häufig in Kanal- und Transportproteinen auf, welche Konformationsänderungen vollführen. In proteingebundenen Liganden und Coenzymen wurde eine zu den Referenzdaten ähnliche Packungsdichte beobachtet. Mit diesen Ergebnissen konnten mehrere Widersprüche in der Fachliteratur ausgeräumt werden. / In this work, a multitude of data on structure and function of proteins is compiled and subsequently applied to the analysis of atomic packing. Structural analyses often require specific protein datasets, based on certain properties of the proteins, such as sequence features, protein folds, or resolution. Compiling such sets using current web resources is tedious because the necessary data are spread over many different databases. To facilitate this task, Columba, an integrated database containing annotation of protein structures was created. Columba integrates sixteen databases, including PDB, KEGG, Swiss-Prot, CATH, SCOP, the Gene Ontology, and ENZYME. The data in Columba revealed that two thirds of the structures in the PDB database are annotated by many other databases. The remaining third is poorly annotated, partially because the according structures have only recently been published, and partially because they are non-protein structures. The Columba database can be searched by a data source-specific web interface at www.columba-db.de. Users can thus quickly select PDB entries of proteins that match the desired criteria. Rules for creating datasets of proteins efficiently have been derived. These rules were applied to create datasets for analyzing the packing of proteins. Packing analysis measures how much space there is between atoms. This indicates regions where a high local mobility of the structure is required, and errors in the structure. In a reference dataset, a high number of atom-sized cavities was found in a region near the protein surface. In a transmembrane protein dataset, these cavities frequently locate in channels and transporters that undergo conformational changes. A dataset of ligands and coenzymes bound to proteins was packed as least as tightly as the reference data. By these results, several contradictions in the literature have been resolved.

Protein sequences contain the information in order for a protein to fold to a unique compact, three-dimensional native structure. The forces that drive protein structures to form compact folds are largely dominated by burial of hydrophobic amino acids, which results in non-specific packing of amino acid side-chains. The knob-socket model attempts to organize side-chain packing into tetrahedral packing motifs. This tetrahedral motif is characterized with a three residues on the same secondary structure forming the base of the tetrahedron packing with a side-chain from a separate secondary structure. The base of the motif is termed the socket, and the other side-chain is called the knob. Here, we focus on extending the knob-socket model to understand tertiary and quaternary structure. First, single knobs sometimes pack into more than one socket in real structures. We focus on understanding the topology and amino acid preferences of these tertiary packing surfaces. The main results from the study of tertiary packing surfaces is that they have a preferred handedness, some interactions are ancillary to the packing interaction, there are specific amino preferences for specific positions in packing surfaces, and there is no relationship between side-chain rotamer of the knob packing into the tertiary packing surface. Next, we examine the application of the knob-socket to irregular and mixed packing in protein structure. The main conclusions from these efforts show canonical packing modes between secondary structures and highlight the important of coil secondary structure in providing many of the knobs for packing. Third, we investigate protein quaternary structure with a clique analysis of side-chain interactions. We identify a possible pseudo knob-socket interaction, and compare knob-socket interactions between tertiary and quaternary structure. Lastly, we discuss the workflow used in CASP12 to predict side-chain contacts and atomic coordinates of proteins.