An Effective Ensemble Deep Learning Framework for Blood-brain Barrier Permeability Prediction

As a highly protective biological structure, the blood-brain barrier prevents the uncontrolled passage of molecules to keep the central nervous system free from chemical toxification and maintain brain homeostasis. Since most substances are not allowed to freely penetrate the blood-brain barrier, examination of the blood-brain barrier permeability (BBBP) of drug candidates is highly essential in drug discovery. To screen the BBBP of molecules, several computational methods were developed with satisfactory outcomes. These methods, however, have shortcomings that need to be addressed to improve prediction performance. In our study, we propose iBBBP-Ensemble, an ensemble deep learning model that combines two types of neural networks: a convolutional neural network and multilayer perceptrons, and three types of molecular representations: the Extended-Connectivity Fingerprint, RDKit molecular descriptors, and Mol2vec-embedded features. Experimental results confirmed the effectiveness and stability of our proposed model. The benchmarking analysis also indicated that iBBBP-Ensemble outperformed all machine learning and deep learning baseline models.