Prediction Of Cross-Species Protein-Protein Interaction Using Deep Learning And Analysis Of Computational Protein Interactome In Tomato

Protein-protein interactions(PPIs)play key role in most biological processes and functions in living cells.The study of PPI is of great important to understand the molecular mechanism underlying cell function.For the past decades,many experimental methods have been developed for PPI discovery,but the experimental methods are often time-consuming and laborious,which makes the experimental discovery of PPIs still limited in a few model organisms.Therefore,the development of a computational method to find the features of PPI dataset in model species,and then apply it to predict PPIs in other horticultural crops,which will accelerate the molecular dissection of important agronomic traits in horticultural crops.In this study,we developed a deep learning-based method,Deep Cro PPI(Deep Learning based Cross-Species Prediction of Protein-Protein Interaction).This method detect the features of amino acid sequence feature and interaction feature respectively,which can accurately predict cross-species PPIs only from sequences.The results showed that the prediction accuracy of Deep Cro PPI is 82.9% to 93.1% on the datasets of six model organisms.Cross-species prediction performance showed that the average accuracy of single-species dataset is 60.7%.In addition,the test with mixed-species datasets showed that the cross-species prediction accuracy is increased by 2.62% on average comparing with single-species dataset.Furthermore,we optimized the ratio of positive and negative samples in model training.When the positive and negative ratio was adjusted to 1:10,the true positive rate(TPR)is 8.9%,while the false positive rate(FPR)is 0.15%.A series of test showed that Deep Cro PPI could achieve cross-species PPIs prediction,and the FPR was under the expected level.Using the Deep Cro PPI,we recognized the interacting protein pairs out from all possible 640 million pairs in Solanum lycopersicum genome.It resulted in the Sly PPINet(Solanum lycopersicum Protein-Protein Interaction Networks),which consisted of 104,398 pairs of protein interactions,involving 8,238 proteins,covering 23.03% of the whole proteome of S.lycopersicum.Compared with the prediction datasets of S.lycopersicum in string database,a total of 7,603 PPIs can be successfully predicted by the two methods(P < 6.76e-254).Furthermore,the proteins interacting with ethylene signaling pathway were analyzed in Sly PPINet and 1,468 candidate proteins were identified to be involved.Function and expression profiles implied that these genes,such as Solyc06g008580,Solyc01g095900,Solyc12g057110 and Solyc11g044560,might be important in ethylene signaling to mediate fruit development and maturity.Deep Cro PPI and Sly PPINet provide a landscape of computational interactome to facilitate better understanding the molecular basis of important agronomic traits and identifying genes for genetic improvement in S.lycopersicum and other horticultural crops.