| This thesis describes computational structure prediction methods I developed for the study of membrane proteins. Protein structure prediction may be considered as two almost independent stages: the modeling of the backbone, followed by the optimization of the side chains for each backbone geometry. Side chain optimization can become the bottleneck stage of structure prediction, and therefore, needs to be as efficient as possible. In the first part of this thesis, I describe novel methods to improve the speed and accuracy of side chain modeling, which I later leverage to predict the structure of membrane protein complexes. Side chain optimization is a highly combinatorial task complicated by the great degree of side chain conformational freedom. A common approach to model side chain flexibility is to discretize the space in a set of representative conformations, called conformer libraries. These libraries need to provide sufficient sampling of the underlying space, while remaining as small as possible, for the sake of computational efficiency. To achieve a good balance between these conflicting needs, I have developed a novel energy-based criterion to create conformer libraries (chapters 2 to 4). Through experiments I demonstrate that these energy-based conformer libraries enable faster and more accurate side chain modeling using a smaller number of conformers. Membrane proteins often associate with each other to form complexes which are essential for their function. I have developed high-throughput methods, enhanced by the use of energy-based conformer libraries, for predicting the structure of these complexes (chapters 5 and 6). The method described in chapter 5 interprets "low resolution" experimental results, such as mutagenesis data, to create a structural model of the bacterial division protein FtsB. This model has been used to guide the experimental characterization of the FtsB protein. Ab initio structural prediction methods are important when experimental results are not available. Chapter 6 describes the geometric analysis of a common transmembrane motif (GASright) which reveals that the motif is optimized for Calpha hydrogen bonding. The analysis led to the creation of "CATM", a method that predicts ab initio the structure of GASright motifs at near atomic resolution. |
