Models for Protein Interactions

Recent advances in high-throughput experimental techniques have given us a wealth of protein interaction data, rich in quantity and variety. While the quantity and variety of data present special challenges for modeling, they also present unique opportunities for gaining insight into the proteome by allowing us to leverage multiple perspectives. We present TOPROC (TOpology of PROtein Complexes), an algorithm for inferring the internal interaction topology of protein complexes from both direct proteinprotein interaction data (e.g., yeast two-hybrid: Y2H) and from protein co-complex data (e.g., affinity purification-mass spectrometry: AP-MS). While other methods have attempted to use both these kinds of data, they usually require that co-complex data first be transformed into pairwise protein-protein interaction data under a ’spoke’ or ’clique’ model, a transformation we do not require because we learn the hidden internal topology. We evaluate TOPROC by combining all available data from eight high-throughput proteomic datasets, resulting in coverage of over 85% of the yeast proteome. TOPROC exhibits significantly better sensitivity and specificity than other algorithms for predicting domain-domain interactions from Y2H and AP-MS data. Additionally, it provides us with an improved view of the global protein-protein interaction network in yeast by reducing both the false positive and false negative error rates of the observed data. Finally, TOPROC elucidates the topology of AP-MS purifications, both recovering known complexes (e.g., Arp2/3 and RNA polymerase II) and suggesting new complexes. In doing so, TOPROC provides us with a global view of how proteins interact with each other via their domains to form macromolecular complexes. ∗much of this work was done with Allister Bernard and Alex Hartemink