In Silico Prediction of Blood-Brain Barrier Permeability of Compounds by Machine Learning and Resampling Methods

摘要

The blood-brain barrier (BBB) as apart of absorption protects the central nervoussystem by separating the brain tissue from the bloodstream. In recent years,BBB permeability has become acritical issue in chemical ADMET prediction, but almostall modelswere built using imbalanced data sets, which caused ahighfalse-positiverate. Therefore, we tried to solve the problem of biased data sets and built areliable clas- sification model with 2358 compounds. Machine learning and resampling methods were used simultaneously for the refine- ment of models with both 2D molecular descriptors and mo- lecular fingerprints to represent the chemicals. Through a series of evaluation,werealized that resampling methods such as SyntheticMinority Oversampling Technique(SMOTE) and SMOTE+edited nearest neighbor could effectively solve the problemofimbalanced data sets and that MACCS fingerprint combined with support vector machine performed the best. After the final construction of aconsensus model,the overall accuracy rate was increased to 0.966 for the final external data set. Also, the accuracy rate of the model for the test set was 0.919, with an excellent balancedcapacity of 0.925 (sensitivity) to predict BBB-positive compounds and of 0.899 (specificity) to predictBBB-negative compounds. Compared with other BBB classification models, our modelsreduced the rate of false pos- itives and were more robust in prediction of BBB-positive as well as BBB-negative compounds, which would be quite help- ful in early drug discovery.