UCSF

Evolution is the uniting concept of biology and life. At its fundamental level, it operates by sampling amino acid residues in proteins to optimize stability and function. One definition of the function of a protein is through its interactions with other molecules, especially other proteins. Growing evidence suggests a widespread phenomenon involving the domain of one protein interacting with a short, linearly extended peptide region on another protein, accounting for up to 40% of all protein interactions in the cell. As such, these interactions are manipulated by invasive organisms and human diseases to cause pathogenesis, and are targets for new classes of drugs. Identifying specific interactions is therefore critical for human health. Experimental techniques have characterized thousands of peptide binding events, but conducting experiments can be costly and time-consuming, and their results can be prone to false positives. To both help guide experiments and to analyze their output, there is a need to develop accurate computational methods for predicting protein-peptide interaction specificity. This dissertation addresses this challenge, describing four complementary approaches: (i) a machine-learning algorithm to predict proteolytic cleavage in substrates of pro-apoptotic proteases; (ii) a peptide docking method that models the conformation of peptides in complex with protein binding sites; (iii) statistical analysis of peptide datasets derived from high throughput proteomic experiments to characterize factors mediating binding specificity; and (iv) prediction of peptide interactions contributing to pathogenic invasion.