Modeling of tertiary and quaternary protein structures by homology

The structure of a protein is crucial to understand its function. Despite this importance,
experimentally solved structures are only available for a small portion of the currently known protein
sequences. Comparative or homology modeling is currently the most powerful method used in order
to predict the structure from sequence by the use of homologous template structures. Models,
hence, need to be accurate regarding their three-dimensional coordinates and must represent the
biological active state of the target protein in order to be useful for scientists.
Four goals are pursued in this work in this area of research. Firstly, we increase the coverage of
homology modeling by introducing a method which is able to identify and align evolutionary distant
template structures. The resulting template search and selection procedure is hierarchical. Closely
related template structures are identified accurately and efficiently by standard tools.
A computationally more complex method is invoked in order to identify evolutionary more distant
template structures with high precision and accuracy. Integrated into an automated modeling
pipeline, the developed method is competitive compared to other protein structure prediction
methods.
Secondly, the automated modeling pipeline is applied to a large set of protein sequences to increase
the structural coverage of sequence space. The resulting models and associated annotation data are
stored in a relational database and can be accessed online in order to allow scientists to query for
their protein of interest. Efforts are made to update a selected set of sequences regularly by
shortening the update process without losing accuracy. It is found that the structural coverage of
seven proteomes is increased considerably by this large scale modeling approach.
Thirdly, the modeling of quaternary structure is addressed. Significant room for improvement in the
field of quaternary structure prediction is found when assessing the current state-of-the-art methods
in a double blind prediction experiment. Novel similarity measures are therefore developed to
distinguish proteins with different quaternary structure. We further create a template library built of
structures in their previously defined most likely oligomeric state, to extent the concept of homology
modeling towards the prediction of oligomeric protein structures. In order to select template
structures which share the same quaternary structure with the target structure, a variety of
evolutionary and structural features are investigated. It is shown, that using a combination of these
features for the first time predicts the quaternary structure with high accuracy.
Finally, the performances of methods which predict non-folded (intrinsically disordered) protein
segments are assessed. Current issues are addressed in a field of very active research as more and
more proteins are found to be hubs in interaction networks with considerable disordered portions in
their tertiary structure. In general it is found that such methods perform well, even within the limits
of the test set.