The  30, 000 genes of the human genome are speculated to give rise to 106 proteins through a series of post-translational modifications and gene splicing mechanisms. The majority of these proteins interact with other proteins in processes that impact cellular structure and function. A structural description of protein complexes is vital to understand their function and paves the way for understanding complex biological processes. Further, the structure helps understand the intricacies in the binding process and assists in drug-design. The Human Genome Project has allowed identification of the genome and the National Institute of General Medical Sciences’ Protein Structure Initiative has facilitated rapid determination of structures of proteins in the genome. These resources have now allowed for focus of research on the complex processes of protein interactions. Given the structures of the individual entities forming the complex, can we develop algorithms that efficiently determine the structures of the protein complexes?

Existing techniques for protein complex structure determination fall into three categories: experimental, computational, experimental-computational. The traditional experimental method of x-ray crystallography is not widely applicable for complexes due to problems in crystallizing the complexes. Alternative experimental techniques for protein complex structure determination provide sparse and ambiguous information, making it difficult to extract valuable structural information. Computational techniques on the other hand, do not have an experimental basis and can only “predict” but not “determine” complex structures. The third class of techniques apply computational techniques to extract information from experiments and determine structures of protein complexes. Current approaches in this category focus on determining only the best or top best conformations satisfying the data. This is extremely dangerous, especially when the data available is sparse, since near-optimal solutions might also be plausible. This drives the need for inferential techniques that are based on prior knowledge of protein structures and available data.

Existing computational techniques for protein complex structure determination either use a conformation space or configuration space (degree-of-freedom) viewpoint. The advantage of a configuration space approach is that it allows a minimum-parameter representation of all the conformations of the complex. Each protein complex has a fixed number of degrees of freedom, determined by the number of parameters that must be set to uniquely represent the conformation. For example, a complex with two subunits, when the structure of the subunits are known and flexibility is ignored, has six degrees of freedom–three for translation and three for rotation of one subunit with respect to the other. Every possible conformation of the complex can be represented by a point in an n-dimensional space, n being the number of degrees of freedom of the complex. Determining the complex structure, then becomes a search in this n dimensional space. The configuration space view provides a compact representation of conformations and allows us to develop efficient algorithms. Existing configuration-based approaches resort to grid-based sampling techniques to generate feasible conformations and then test their validity. The problem with sampling is that
conformations which correspond to regularly spaced grid points are not necessarily representative of nearby conformations. Valid conformations in between the grid points might be completely ignored. Hence, there is a need for configuration-based algorithms that are complete in their search.


This thesis will show that
A configuration space analysis of protein complexes using experimental data allows for inference of protein complex structures.