The ability to design drugs that disrupt formation of protein-protein interfaces is of particular interest to the pharmaceutical industry due to its promise for opening an entire new range of drug targets, many of which have already been well characterised in terms of their disease causing effect on the human body. Furthermore these interactions can be involved in many processes unique and essential to bacteria and viruses. We show that pockets on protein-protein interface are smaller but more numerous than those of marketed drugs using a pocket fnding algorithm (Q-SiteFinder). We investigate the similarities and differences between several candidate compounds designed to bind and disrupt protein-protein interfaces and compare to those of current marketed drugs designed to bind more traditional protein targets. We ask the further question as to whether it is possible to better identify pockets on a protein surface as likely to be drug binding. We conclude that it is possible to carefully use random forest machine learning techniques to marginally improve these predictions. However, it is extremely diffcult to use simple physical parameters to provide added information as to the maximal affnity that a small-molecule might be able to achieve in a given binding pocket. Further to the above questions we then investigate the hDM2-p53 system which when disrupted can induce apoptosis in many forms of cancer, making it a target of considerable interest to the pharmaceutical industry. Molecular docking is exploited in order to generate likely structural conformations of oligoamide hDM2-p53 inhibitors which can be used as a starting point for molecular dynamics simulations. These simulations using the AMBER/GAFF force feld are then further developed to perform replica-exchange alchemical free energy calculations using the Bennett Acceptance Ratio non-biased estimator. These simulations are in general shown to be very accurate and show promise in generating hypotheses for novel high-affnity oligoamide compounds.