Proteins are involved in numerous functions in the human body, including chemical transport, molecular recognition, and catalysis. To perform their function most proteins must adopt a specific structure (often referred to as the folded structure). A microscopic description of folding is an important prerequisite for elucidating the underlying basis of protein misfolding and rational drug design. However, protein folding occurs on heterogeneous length and time scales, presenting a grand challenge to both experiments and simulations. In computer simulations, challenges are generally mitigated by adopting coarse-grained descriptions of the physical environment, employing enhanced sampling strategies, and improving computing code and hardware. While significant advances have been made in these areas, for numerous systems a large spatiotemporal gap between experiment and simulations still exists, due to the limited time and length scales achieved by simulation, and the inability of many experimental techniques to probe fast motions and short distances.

In this thesis, kinetic transition networks (KTNs) are constructed for various protein folding systems, via approaches based on the potential energy landscape (PEL) framework. By applying geometry optimisation techniques, the PEL is discretised into stationary points (i.e.~low-energy minima and the transition states that connect them). Essentially, minima characterise the low-lying regions of the PEL (thermodynamics) and transition states encode the motion between these regions (dynamics). Principles from statistical mechanics and unimolecular rate theory may then be employed to derive free energy surfaces and folding rates, respectively, from the KTN. Furthermore, the PEL framework can take advantage of parallel and distributed computing, since stationary points from separate simulations can be easily integrated into one KTN. Moreover, the use of geometry optimisation facilitates greater conformational sampling than conventional techniques based on molecular dynamics. Accordingly, this framework presents an appealing means of probing complex processes, such as protein folding. In this dissertation, we demonstrate the application of state-of-the-art theory, combining PEL analysis and KTNs to three diverse protein systems.

First, to improve the efficiency of protein folding simulations, the intrinsic rigidity of proteins is exploited by implementing a local rigid body (LRB) approach. The LRB approach effectively integrates out irrelevant degrees of freedom from the geometry optimisation procedure and further accelerates conformational sampling. The effects of this approach on the underlying PEL are analysed in a systematic fashion for a model protein (tryptophan zipper,1). We demonstrate that conservative local rigidification can reproduce the thermodynamic and dynamic properties for the model protein.

Next, the PEL framework is employed to model large-scale conformational changes in proteins, which have conventionally been difficult to probe \textit{in silico}. Methods based on geometry optimisation have proved useful in overcoming the broken ergodicity issue, which is associated with proteins that switch morphology. The latest PEL-based approaches are utilised to investigate the most extreme case of fold-switching found in the literature:~the $\alpha$-helical hairpin to $\beta$-barrel transition of the C-terminal domain of RfaH, a bacterial transcription factor. PEL techniques are employed to construct the free energy landscape (FEL) for the refolding process and to discover mechanistic details of the transition at an atomistic level.

The final part of the thesis focuses on modelling intrinsically disordered proteins (IDPs). Due to their inherent structural plasticity, IDPs are generally difficult to characterise, both experimentally and via simulations. An approach for studying IDPs within the PEL framework is implemented and tested with various contemporary potential energy functions. The cytoplasmic tail of the human cluster of differentiation 4 (CD4), implicated in HIV-1 infection, is characterised. Metastable states identified on the FEL help to unify, and are consistent with, several earlier predictions.