| Knowing which proteins interact is essential for learning about biological processes and functions, as almost every process in the cell involves a protein interaction. Determining protein-protein interactions is one of the most well-studied yet difficult problems in biology. The large amount of genomic data---20,000+ genes in human alone---and the combinatorial number of possible interactions make it impossible to test all proteins in the laboratory in a short time, even using high-throughput techniques, and motivates the study of efficient computational methods. We focus on protein interactions mediated by two-stranded, parallel coiled-coil, one of the most common structural motifs. Coiled coils are involved in many cellular processes, including transcription, oncogenesis, cell structure, and cell fusion events. We present several computational methods for predicting coiled-coil protein interactions.; First, we describe a learning method that represents coiled coils in terms of their interhelical interactions and derives, from sequence and experimental data, a weight that indicates how favorable each residue-residue interaction is. When tested on interactions between human and yeast bZIP transcription factor proteins, our method identifies 70% of strong interactions while maintaining that 92% of predictions are correct. Showing its usefulness outside this protein class, we apply our method to predict novel bZIP interactions in Arabidopsis and to computationally identify all coiled-coil interactions in the yeast proteome. While unable to identify coiled coils with high confidence at the proteomic scale, our method makes inroads towards this more challenging problem. Finally, we integrated three successful predictors using a probabilistic framework. The combined method is demonstrated to perform better than any individual method, on average.; Our work is the first to predict protein-protein interactions with such high confidence. While developed for coiled coils, the techniques can also be applied towards other structural domains and general protein-protein interactions. |
