X-Ray structure re-refinement. Combining old data with new methods for better structural bioinformatics.

Nearly all processes in living organisms are governed by proteins. Research questions in (medical) biology can therefore often be reduced to questions about the biochemistry of a particular protein. E.g.: “Why does Tamiflu not work against a novel influenza mutant and how should the drug be adapted to become effective again?”. Answering such questions requires an accurate and detailed structural description of the protein under investigation in which each individual protein atom has three-dimensional (3D) coordinates. These 3D structures of proteins are typically determined using X-ray crystallography which is difficult and labour intensive. As a result, we know far fewer protein structures than actual proteins. To advance our scientific knowledge we have to make do with the structures currently available. In this thesis three bioinformatics approaches to get the most out of current structures are explored. The first approach is structure validation which uses (statistical) knowledge of molecules and chemical interactions to detect good, bad, and ugly parts of protein structures. This helps to select the optimal protein structure to answer specific research questions and helps to estimate the reliability of scientific conclusions drawn from protein structures. The second approach is improving the quality of current structures. Not by completely repeating the X-ray crystallography experiments, but by using the original experimental data and the latest computational methods. We explore this in depth with a proof-of-concept experiment followed by a high throughput computational experiment on 17,000 structures and an evaluation of what structural elements can be improved by actively fixing errors in structures. The third and most constructive approach is making new protein structures via homology modelling. We discuss the importance of combining wet-lab and computational protein science and show that homology modelling depends on high-quality protein structures as a starting point. This stresses the importance of improving the current structures