Prediction Of Blood-brain Barrier Peptides Based On Machine Learning

Neurological and mental diseases are one of the main factors of disability,and there is a lack of effective ways to treat these diseases.This is because almost all of macromolecular neuro-therapies and over 98% of small-molecule drugs are blocked by the physical and enzyme barrier called the blood-brain barrier(BBB).Compared with other therapeutic drugs and methods,blood-brain barrier peptides(BBP)has the advantages of being amenable for chemical synthesis and low toxicity.BBPs play a host of direct or indirect central nervous system effects and frequently exhibit hormesis,and they open the new diagnostic and therapeutic avenues for brain diseases.In order to minimize the cost of experiment and improve the experimental efficiency,it is a need to develop efficient computational approaches for predicting BBPs.In this paper,we proposed a methods to predict the blood-brain barrier peptide,and further improved the performance of the tool through the feature representation learning scheme.The research contents are divided into the following two aspects:1.We develop a novel BBP prediction method for identify BBP based on s biological features.First,we construct a dataset of BBP through public databases and literature.Subsequently,a series of properties,such as molecular weight length,lipophilicity and electric charge,were analyzed by literature investigation.Based on these characteristic,we encode a set of biological features of BBP.Finally,we evaluate model based on well-prepared up-to-date training and independent test datasets.The experimental results show that our prediction method achieves comprehensive performance on the training set and the independent data set,and the AUC of the test set reaches 0.8255.2.We present a sequence-based prediction approach called BBPpred(Blood-brain Barrier Peptides Prediction),that can efficiently identify BBPs using logistic regression.We firstly use a feature representation learning scheme that learns the most discriminative features from existing feature descriptors in a supervised way.To improve the feature representation ability,seven informative features are selected for classification by removing noisy and irrelevant features.In addition,we specifically create two benchmark datasets(training and independent test),which contain a total of 119 BBPs from public database and the literature.In 10-fold cross-validation on the training dataset,BBPpred achieves promising performances with an AUC score of 0.8764 and an AUPR of 0.8757.We also test our developed method on the independent test dataset and gain a competitive advantage.BBPpred proposed in this paper is the first prediction tool of BBP so far.The research content of this paper can provide theoretical guidance for us to characterize the functional mechanism of BBP and research and development of drugs related to nervous system diseases.