A bioinformatics study of protein folding, aggregation and disease

The objective of this PhD thesis is to provide insightful analyses and biological data resources of protein structural and sequence features that are strongly associated with protein function and disease. In terms of protein sequence features, my research focuses on data collection, analysis and knowledgebase construction in order to allow the generation of new hypothesis on protein functions and follow-up studies of human diseases. Protein structure and disorder are introduced in Chapter 1. Chapter 2 focuses on analysing protein sequence features of human polyglutamine (polyQ) proteins that contain consecutive glutamine repeats in their sequences. A number of studies have demonstrated that expanded polyQ repeats are responsible for human neurodegenerative disease including Huntington disease and spinocerebellar ataxia. Building upon the previously published PolyQ database, an updated database, named PolyQ 2.0, has been constructed by incorporating functional and structural annotations for human disease- and non-disease associated polyQ proteins. Chapter 3 describes a novel knowledge-base, ‘KinetochoreDB’, a relational database that describes kinetochore and its related proteins. Kinetochore plays a crucial role during cell mitosis and meiosis by pulling sister chromatids apart. A number of disease-associated mutations have been verified and located with the kinetochore and its related proteins. It is envisaged that this new database will be useful for studies of kinetochore proteins and related diseases. From a protein structure perspective, two important structural features have been investigated: protein coiled-coil domains (CCDs) and disordered regions. CCD is a type of protein tertiary structure consisting of an ensemble of helices binding together. It has been estimated that approximately 10% of eukaryotic proteins contain CCDs. Previous reports have revealed that some mutations occurring within the CCDs are responsible for human diseases due to the instability of CCDs caused by these mutations. On the other hand, CCDs have been widely used as drug delivery systems due to their capability for molecular binding and recognition. From a bioinformatics perspective, this thesis mainly focused on analysing computational approaches for accurate identification of CCDs and their oligomeric states. Chapter 4 presents a comprehensive and critical performance evaluation of 12 currently available computational approaches for protein CCD and oligomeric state prediction, using carefully curated independent test datasets. A study of nine human polyQ disease-associated proteins was also performed to illustrate the prediction inconsistency amongst different CCD and oligomeric state predictors and highlighted the useful directions for development of improved predictors. Intrinsically disordered proteins (IDPs) lack stable and well-defined three-dimensional structures. Proteins with disordered regions are often biologically important and play crucial roles in molecular binding and recognition, protein post-translational modification, protein regulation and other important biological processes. Using well-annotated datasets of human disease-associated mutations and state-of-art computational algorithms for protein disorder prediction, I investigated four different types of structural transitions due to single point mutations, all underlying human pathogenic mutations and polymorphisms; Disorder-to-Order (D→O), Disorder-to-Disorder (D→D), Order-to-Disorder (O→D) and Order-to-Order (O→O). Chapter 5 presents a bioinformatics analysis of these four structural transitions and proposes a mechanism, named ‘structural capacitance’ that may lead to de novo generation of microstructure in previously disordered regions (for D→O structural transition). A summary, discussion and future directions of all the topics covered in this thesis are provided in Chapter 6.