Computer-aided drug discovery has become an important component in the development of new therapeutics. Despite significant progress in software and hardware there are still a multitude of unsolved issues that limit the accuracy of these computational methods. Our research has been dedicated to the development and application of computational methods addressing serious shortcomings of present computational approaches for modeling protein-ligand and protein-protein association. In this talk, I will discuss a few examples of our recent methodological developments and their applications to rational lead optimization, binding affinity predictions and prevent the fibrillation of peptide drugs. In particular, I will discuss our software WATsite for predicting the influence of water on protein-ligand binding, Limoc for incorporating protein flexibility into protein-ligand docking, and IterTunnel to model kinetic effects of ligand binding. Finally, I will highlight our efforts on predicting fibrillation of peptides using Fibpredictor and the rational design of glucagon prodrugs to prevent aggregation.