Predicting the Brain-To-Plasma Unbound Partition Coefficient of Compounds via Formula-Guided Network

Blood-brain barrier (BBB) permeability plays a crucial role in determining drug efficacy in the brain, with the brain-to-plasma unbound partition coefficient (<i>K</i>_p,uu) recognized as a key parameter of BBB permeability in drug development. However, <i>K</i>_p,uu data are scarce and mostly in-house. In predicting <i>K</i>_p,uu the generality and applicability of existing empirical scoring models remain underexplored. To address this, we established a public rat <i>K</i>_p,uu data set through data mining and developed a formula-guided deep learning model, CMD-FGKpuu, which performed well on multiple benchmark tests, marking good demonstration of the potential of deep learning for <i>K</i>_p,uu prediction. Additionally, the model can be fine-tuning with project-specific experimental data, thus improving its practical utility. The findings offer an effective tool for predicting BBB permeability in drug development and introduce a new perspective for applying few-shot learning in the pharmaceutical field.