Decoding the Blood-Brain Barrier: Innovative and Scalable Open-Source Machine Learning Model for Drug Permeability.

INTRODUCTION: The permeability of the blood-brain barrier (BBB) is an important property for potential drugs. Various in vitro and in vivo methods have been established to measure bloodbrain transport. Nevertheless, empirical assessment of the blood-brain barrier (BBB) for all drug candidates is both resource-intensive and costly. METHODS: In this study, we present practical and reliable machine learning methods developed using extensive datasets to assess BBB permeability. RESULTS: The best model for the BBB classification task reached the ROC-AUCcv value of 0.963, whereas our best BBB regression model achieved R2 = 0.954, Q2 = 0.728, and RMSEcv = 0.321. DISCUSSION: The study introduced the novel approach of using classification labels as additional descriptors for regression tasks, which significantly improved model performance. The models demonstrated strong generalization capabilities, with validation metrics closely matching crossvalidation results. CONCLUSION: Due to the significantly expanded dimensions of the training dataset collected from a variety of sources, the regression model we developed exhibits greater robustness compared to previously published models. We also demonstrated for the first time the idea of using a classification label as an additional descriptor for a regression task. Among the most important things is the fact that all our models are publicly available and can be used by scientists for their own research.