From protein-protein interactions to rational drug design: are computational methods up to the challenge?

The study of protein-protein interactions (PPIs) has been growing for some years now, mainly as a result of easy access to high-throughput experimental data. Several computational approaches have been presented throughout the years as means to infer PPIs not only within the same species, but also between different species (e.g., host-pathogen interactions). The importance of unveiling the human protein interaction network is undeniable, particularly in the biological, biomedical and pharmacological research areas. Even though protein interaction networks evolve over time and can suffer spontaneous alterations, occasional shifts are often associated with disease conditions. These disorders may be caused by external pathogens, such as bacteria and viruses, or by intrinsic factors, such as auto-immune disorders and neurological impairment. Therefore, having the knowledge of how proteins interact with each other will provide a great opportunity to understand pathogenesis mechanisms, and subsequently support the development of drugs focused on very specific disease pathways and re-targeting already commercialized drugs to new gene products. Computational methods for PPI prediction have been highlighted as an interesting option for interactome mapping. In this paper we review the techniques and strategies used for both experimental identification and computational inference of PPIs. We will then discuss how this knowledge can be used to create protein interaction networks (PINs) and the various methodologies applied to characterize and predict the so-called "disease genes" and "disease networks". This will be followed by an overview of the strategies employed to predict drug targets.