An Enhanced Prediction Framework for Blood-Brain Barrier Permeability: DeePred-BBB

Introduction: The blood-brain barrier (BBB) is a semipermeable, discerning barrier that keeps the CNS's internal environment stable. Constructing a concise statement on BBB permeability is challenging, yet it remains a crucial factor in developing central nervous system (CNS)-acting drugs. Clinical studies provide the most reliable assessment of BBB permeability, but they require substantial time and effort. Consequently, various computational approaches have been explored to estimate BBB permeability. Methods: However, there has always been a problem with the precision of models used to predict BBB permeability. Using a dataset of 3912 chemicals, we trained a deep-learning (DL) and machine learning (ML) algorithm to better predict BBB permeability. There were 1,917 features stored for each compound; they included 1356 physicochemical (1D & 2D) attributes, 174 molecular-access-system (MACCS), & 311 substructure fingerprints. Results and Discussion: We compared and contrasted the created models' prediction performance metrics. It was determined that the prediction accuracy of the DNN was 99.58%, the one-dimensional convolutional neural network (CNN-1D) was 98.36%, and the CNN by transfer learning (VGG16) achieved 97.23%. The “DeePred-BBB” framework, which predicts the BBB-permeability of substances utilizing their simplified-molecular-input-line-entry-system (SMILES) notations, was built using the top-performing DNN-based model. It might be helpful in the early phases of medication development for screening compounds according to their BBB permeability. Conclusion: This advancement highlights its potential utility in early CNS drug development. Additionally, “DeePred-BBB” integrates a user-friendly interface, enabling seamless compound screening for researchers and drug developers.