Modeling And Analysis Of Arabidopsis Protein-protein Interaction Network

Study of protein-protein interactions(PPIs)could uncover unknown functions of proteins at the molecular level and gain insight into cellular activities such as growth and development,metabolism,differentiation as well as apoptosis.Currently,although high-throughput experimental techniques have been developed for predicting PPIs,these methods remain resource-consuming and labor-intensive.Large-scale analysis of protein interactions is still far from being popularized in botanical research.As an effective complement to experimental methods,a number of complementary computational approaches have been paid more attention on prediction of PPIs based on genomic context.Based on the previous research,we present an updated version of Arabidopsis protein-protein interaction network(AraPPINet)that takes advantage of increasing information and the introducing of a method to solve the training data imbalance problem.The data resources were improved by new published datasets compared to the previous network,such as gene co-expression analysis using RNA-seq data,of which the proportion of all possible interaction was greatly increased from 59.51% to 96.02%.Moreover,the coverage of structural information was increased from 23.94% to 11.48%.In addition,the predictability of new AraPPINet network is benefited from the reduced effects of class imbalance by using SMOTE technique.The updated AraPPINet contains 345,006 pairs of predicted protein pairs involving 13,929 proteins,with increased accuracy of 8.9% and 10.8%,respectively,compared to previous AraPPINet.Ten-fold cross-validation method was used to measure the performance of the updated AraPPINet.The evaluation shows that the TPR of this method reaches 49.74%,while the FPR remains 0.095%.Precision-recall curves shows that the current method has a better prediction ability than the previous version.In addition,two test datasets of experimentally determined interactions were used to further evaluate the accuracy of the updated AraPPINet.2,252(15.2%)of high-throughput PPIs and 2,880(13.5%)of the new released PPIs could be successfully recognized by the updated network,which significantly overperformed the previous version as well as three other PPI predict methods(AtPID,AtPIN and PAIR).The evaluation results reveal that the prediction accuracy of the updated AraPPINet is significantly improved compared to the previous version.Inferred from the updated version of AraPPINet,we constructed the signaling networks of five hormones including gibberellin,auxin,cytokinin,ethylene and abscisic acid.The updated AraPPINet shows great improvement for predicting PPIs in signaling pathways over the previous version on both two test datasets.For IAA signaling,75% of PPIs identified from high-throughput experiments and 61% of newly reported PPIs could be successfully predicted by the updated network,that was much more accuracy than the previous version(53% and 43%).The five signaling networks were subjected to further evaluations that demonstrated its ability to detect PPIs involved in crosstalk of hormone signaling.Statistical analysis shows that a large amount of proteins interacting with two or more hormonal signaling pathways,which indicates that these shared proteins might play important roles in connecting hormone signaling pathways in plants.The updated version of AraPPINet provides a more reliable interactome which facilitates discovering the crosstalk of the molecular regulation mechanisms of hormone signaling pathways in plants.