The Predicted Arabidopsis Interactome

Knowledge of molecular interaction network is essential in understanding biological function and regulation. Many high throughput experiments have been used to explore the interactome in several model organisms, such as yeast, fly, worm and human. And many major repositories have been established to collect and organize reported interactions. However, even for the model organism in plants, Arabidopsis thaliana, lacking of protein-protein interactions is still a serious issue. Therefore, in this work, we integrated multiple types of indirect evidences, such as gene co-expression, domain interaction, co-localization, annotation similarity, interologs, and co-evolution, into support vector machine models, which were used to predict interactions for Arabidopsis. This resulted in 145494 potential interactions. Furthermore, we created a database (PAIR, http://www.cls.zju.edu.cn/pair/) hosting our predicted interactions and experimentally reported interactions. This database features rich annotations and an interactive network browser. Lastly, we demonstrated the usefulness of our predicted interactions by analyzing the meiosis recombination-related protein network.