Understanding genetic variation is the basis for prevention and diagnosis of inherited disease. In the `next generation sequencing' era with rapidly accumulating variation data, the focus has shifted from population-level analyses to individuals. This thesis is centred on the problem of gathering, storing and analysing mutation data to understand and predict the effects single amino acid mutations will have on protein structure and function. I present analysis of a subset of mutations and a new predictive method implemented to expand the coverage of the structural effects by our pipeline. I characterised a subset of pathogenic mutations: `compensated pathogenic deviations'. These are mutations which cause disease in humans, but the mutant residues are found as native residues in other species. During evolution, they are presumed to spread through populations by coevolving with another, neutralising mutation. When compared with uncompensated mutations, they often cause milder structural disruptions, prefer less conserved structural environments and are often found on the protein surface. I describe the development of a new analysis to test the eeffects of mutations by predicting residues involved in protein-protein interfaces where the structure of the complex is unknown. Two machine learning methods (multilayer perceptrons and, in particular, random forests) show an improvement over previously published protein-protein interface predictors. This new method further increases the ability of the SAAPdb analysis pipeline to show the effects of mutations on protein structure and function. Furthermore, it is a template for building prediction-based structural analysis methods for the pipeline, where available structural data are insufficient. In summary this thesis examines mutations from both an evolutionary and a disease perspective. In addition, a novel method for predicting protein interaction regions is developed thus expanding the existing pipeline and furthering our ability to understand mutations and use them in a predictive context.