Construction And Prediction Of Blood-brain Barrier Penetrating Peptide Databas

The blood-brain barrier is the main obstacle for drugs to enter the brain parenchyma.Blood-brain barrier penetrating peptides are a class of bioactive peptides that can non-invasively pass through the blood-brain barrier.They can be used as carriers for brain delivery of drugs for central nervous system diseases.In the development of drugs for central nervous system diseases,there are important research significance and potential applications prospect in these peptides.To promote research on blood-brain barrier penetrating peptides,a database dedicated to storing detailed information of blood-brain barrier penetrating peptides and a prediction tool that can quickly and accurately identify blood-brain barrier penetrating peptides were built in this thesis.By fully mining the relevant information of blood-brain barrier penetrating peptides and collecting them experimental related data,a comprehensive blood-brain barrier penetrating peptide database named as BBPDB was established(http://i.uestc.edu.cn /bbpdb/).The database provides 24 fields of information for each blood-brain barrier penetrating peptide,such as peptide sequence,the type of disease targeted et.al.At present,the BBPDB database contains a total of 651 blood-brain barrier penetrating peptides,of which 247 peptides have defined the types of diseases they target,and four peptides have entered the clinical research stage.BBPDB database can provide data support for the study of blood-brain barrier penetrating peptides.Based on pure sequence information,a web page predictor named as BBPpredict was proposed to quickly and freely identify blood-brain barrier penetrating peptides(http://i.uestc.edu.cn/BBPpredict/cgi-bin/BBPpredict.pl).This predictor was constructed based on 326 strictly screened blood-brain barrier penetrating peptides and326 non-blood-brain barrier penetrating peptides.Five feature code methods such as amino acid composition and pseudo-amino acid composition were used to extract the internal information of peptides,F-Score feature scoring algorithm and grid search strategy were utilized for selecting optimal features and algorithm parameters.Finally,the prediction performance of different machine learning algorithms and deep learning algorithms was evaluated using five-fold cross-validation and an independent test dataset.The random forest algorithm was chosen to build the final classification model,and this model achieved an accuracy of 77.27% on the independent test dataset.Compared with the existing prediction tools,its prediction performance has been significantly improved.It can be used as a preliminary screening tool for the verification of blood-brain barrier penetrating peptides.