Prediction And Study Of “in Situ Crosstalk” Sites Of Serine Phosphorylation And ADP-ribosylation Modification Based On Ensemble Learnin

Protein Translational Modifications（PTMs）regulate Protein function and activity,and different post-translational modifications enable proteins to play different roles in the intracellular Protein interaction network.At present,it is found that amino acids at the same site of proteins can be modified by different kinds of proteins after translation,which is called"situ crosstalk"."Crosstalk"adds specificity and combinatorial logic to signal processing,but has not been studied on a large scale,so"crosstalk"is an important topic in biology.Serine ADP-ribosylation（SADPr）modification is closely related to DNA damage.It controls the state of the cell’s chromatin to ensure efficient DNA repair.In addition,serine phosphorylation（pS）is a common and important post-translational protein modification in biology,which is closely related to DNA damage repair,gene transcription regulation,signal cell transduction,apoptosis regulation and other biological processes.In this study,we investigated the characteristics of pS and SADPr"situ crosstalk"（pSADPr）sites occurring at the same site,constructed a prediction algorithm for pSADPr sites,and compared the similarities and differences between pSADPr sites and single type modification sites.In this study,we investigated the characteristics of pS and SADPr"situ cross-talk"（pSADPr）sites occurring at the same site,constructed pSADPr site prediction algorithm,compared the similarities and differences between pSADPr sites and a single type of modification site,and developed an efficient and universal pSADPr modification site prediction model.The discussion and analysis involved include the following three points:（1）In this study,3250 pSADPr sites were collected in a large scale for the first time,which were located on 1360 proteins.Data were collected from the intersection of 7,520SADPr modification sites and 151,227 pS modification sites in humans.（2）Based on the sequence information characteristics of pSADPr sites,an integrated deep learning model EdeepSADPr based on stacking integrated learning principle is constructed for the first time to predict pSADPr sites.With stacking integrated learning algorithm,one-dimensional convolutional neural network（CNN）is combined with one-hot,ZSCALE and Word Embedding to build three single classification models CNN_OH,CNN_ZSCALEand CNN_WE as base models.The predicted probability values constitute new feature vectors,and then the random forest algorithm is used as a meta-model for training,and the final model is obtained（3）In this study,three benchmark data sets were constructed for different situations:1)Serine sites known to be experimentally verified were modified by ADP-ribosylation to predict whether this site was a pSADPr site.pSADPr sites were taken as positive samples and single modified SADPr sites were taken as negative samples for modeling training.2)It is known that the serine site verified by experiments is phosphorylated and modified,and it can be predicted whether this site is pSADPr site.pSADPr sites were taken as positive samples and single modified pS sites as negative samples for modeling training.3)Predict whether a serine site（Us）with unknown modification status is a pSADPr site.The pSADPr site was taken as a positive sample and the Us site as a negative sample for modeling training.The AUC values of EdeepSADPr model corresponding to the three groups of benchmark data sets are 0.795,0.917 and 0.944,respectively,which are better than the models built by other algorithms.In conclusion,EdeepSADPr is an efficient and robust classifier.