Multi-dataset electron density analysis methods for X-ray crystallography

Thesis

X-ray crystallography is extensively deployed to determine the structure of proteins, both unbound and bound to different molecules. Crystallography has the power to visually reveal the binding of small molecules, assisting in their development in structure-based lead design. Currently, however, the methods used to detect binding, and the subjectivity of inexperienced modellers, are a weak-point in the field.

Existing methods for ligand identification are fundamentally flawed when identifying partially-occupied states in crystallographic datasets; the ambiguity of conventional electron density maps, which present a superposition of multiple states, prevents robust ligand identification. In this thesis, I present novel methods to clearly identify bound ligands and other changed states in the case where multiple crystallographic datasets are available, such as in crystallographic fragment screening experiments. By applying statistical methods to signal identification, more crystallographic binders are detected than by state-of-the-art conventional approaches.

Standard modelling practice is further challenged regarding the modelling of multiple chemical states in crystallography. The pervading modelling approach is to model only the bound state of the protein; I show that modelling an ensemble of bound and unbound states leads to better models. I conclude with a discussion of possible future applications of multi-datasets methods in X-ray crystallography, including the robust identification of conformational heterogeneity in protein structures.

Authors: Pearce, N

• DOI: 10.5287/ora-8npzq055g
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: PanDDA
• Subjects: Data Analysis Methods; X-ray Crystallography