The blood–brain barrier (BBB) plays a central role in maintaining hormonal homeostasis between the circulation and the brain. Many endocrine-active compounds—such as steroid hormones, peptide analogs, and metabolic modulators—must cross or avoid the BBB to achieve their intended therapeutic effects. Experimental BBB assays are costly and limited in throughput. This study aims to develop and validate machine learning (ML) models for predicting BBB permeability of endocrine-related molecules, thereby accelerating the discovery of safer and more effective neuroendocrine therapeutics.

A curated dataset of 600 compounds with experimentally defined BBB permeability labels was analyzed. Physicochemical and structural descriptors were generated using PaDEL and RDKit. Five ML algorithms—logistic regression, support vector machine (SVM), random forest (RF), AdaBoost, and XGBoost—were trained with stratified 10-fold cross-validation and optimized by grid search. Model performance was evaluated using accuracy, F1-score, and the area under the ROC curve (AUC). The best model was further tested on an independent external dataset to assess generalizability.

Among all models, the ensemble-based XGBoost classifier exhibited the best performance, achieving an AUC of 0.818 in internal cross-validation and 0.839 in external validation. It outperformed AdaBoost (0.772), RF (0.743), SVM (0.609), and logistic regression (0.579). Feature importance analysis identified lipophilicity, molecular weight, and topological polar surface area as key determinants of BBB penetration. The consistent internal and external performance demonstrates high model robustness and reproducibility across diverse chemical scaffolds relevant to endocrine pharmacology.

This study presents a reliable, externally validated ML framework for predicting BBB permeability of endocrine-active compounds. The XGBoost model shows strong potential as an early screening tool to optimize CNS exposure profiles and minimize off-target neuroendocrine effects. Integrating such predictive modeling into preclinical pipelines can accelerate the design of safer hormone-based and neuroendocrine therapeutics. Future work will extend to deep molecular embeddings and integrative models coupling BBB transport with hormonal receptor activity for precision endocrinology.