Peptide Feature Extraction And New Functional Peptide Discovery For Vascular Stents Coating

Implantation of vascular stents is an important means of treating coronary heart disease,but the implantation itself may also bring serious risks such as vascular restenosis and thrombosis afterwards.Modification of stent surfaces with functional peptides is an effective way to reduce the associated risks.However,the functional peptides currently available for surface modification of vascular stents are quite limited,and their performance also has certain limitations.To this end,based on the published data of human blood proteins and extracellular matrix proteins,this study built a specialized peptide database for the development of vascular stent coating materials,investigated the characteristics of exposed peptides which were located on the surface of proteins,and adopted artificial intelligence methods to establish a generative model for generating peptides similar to exposed peptides.In addition,the study also used the word vector approach to obtain several novel functional peptides which had similar functions to the target peptide.Focused on the published human blood protein and extracellular matrix protein data,the study first established a specialized peptide database and analyzed the position of tripeptides of each protein in the database.The tripeptides were divided into exposed peptides and buried peptides.The support vector machine method was used for the classification of two kinds of peptides,and classifiers were trained and retained.According to the results of the classifiers,the study further analyzed the amino acid composition and physicochemical properties of tripeptides.The results showed that the exposed peptides tended to have amino acids like G,E,D,K,but not C,W,M.The study investigated the possibility of using recurrent neural networks to generate peptides with similar properties to peptides located on the surface of proteins.Based on the sequence information of exposed peptides from human blood proteins and extracellular matrix proteins in the specialized database,two generative models were trained which could de novo generate novel peptides of any length that were similar to the exposed peptides.Peptides generated by the models were analyzed by the corresponding classifiers.The results showed that 78.6%and 89.7%of the tripeptides generated by the two models were identified as exposed peptides,respectively.As a control,the proportions of tripeptides identified as exposed peptides which were generated by the random method were 63.6%and 58.8%.In addition,the Euclidean distances between model-generated peptides and known exposed peptides,and between randomly generated peptides and known exposed peptides were also calculated,of which the former was smaller than the latter.The results indicated that the modelgenerated peptides were more similar to the known exposed peptides in terms of biochemical properties.Finally,the study also attempted new methods of finding novel peptides similar to a given functional peptide from the specialized database.Using the specialized database as the corpus,word vectors were established with Word2Vec,GloVe,and BERT models.According to the results of word vector similarity,Word2Vec and GloVe models output 10 novel peptides which were similar to peptide RGD from nearly 8000 candidates respectively.Molecular simulation of the results of the Word2Vec model showed that 6 out of 10 peptides had similar or better effects on endothelial cell adhesion promotion than the target peptide RGD.The conclusion was also supported by the experimental results from the collaborative team.Experiments showed that all the 10 novel peptides could improve the adhesion of endothelial cells,of which two were comparable to RGD.