Masked graph transformer for blood-brain barrier permeability prediction.

Area of Science:

• Computational Chemistry
• Neuroscience
• Drug Discovery

Background:

• The blood-brain barrier (BBB) protects the central nervous system but hinders drug delivery.
• Accurate assessment of blood-brain barrier permeability (BBBP) is crucial for identifying viable drug candidates.
• Existing computational methods for BBBP prediction have limitations in predictive accuracy.

Purpose of the Study:

• To develop and evaluate novel classification models for predicting BBBP.
• To leverage a Masked Graph Transformer-based Pretrained (MGTP) encoder for molecular featurization.
• To improve the generalizability and predictive power of computational BBBP screening tools.

Main Methods:

• Constructed classification models using chemical data featurized by an MGTP encoder.
• Employed masked attention-based learning to train the MGTP encoder for enhanced molecular structure encoding.
• Evaluated model performance against existing representations across multiple datasets.

Main Results:

• Classification models utilizing MGTP features outperformed models with other representations in 6 out of 8 cases.
• The MGTP encoder demonstrated effectiveness in generating discriminative molecular features.
• Chemical diversity analysis validated the encoder's capability to differentiate molecular classes.

Conclusions:

• The proposed MGTP encoder significantly enhances the prediction of blood-brain barrier permeability.
• This approach offers a more effective computational tool for screening drug candidates in early-stage drug discovery.
• The MGTP encoder shows promise for various downstream tasks requiring robust molecular representation.