Evaluating the Use of Graph Neural Networks and Transfer Learning for Oral Bioavailability Prediction.

Area of Science:

• Pharmacokinetics
• Computational Chemistry
• Drug Discovery

Background:

• Oral bioavailability is crucial for drug discovery, influencing therapeutic efficacy.
• Traditional computational models rely on manual feature selection, requiring domain expertise and time.
• Graph neural networks (GNNs) offer automated feature extraction capabilities.

Purpose of the Study:

• To predict oral bioavailability using GNNs with automated feature selection.
• To enhance GNN performance for oral bioavailability prediction through transfer learning.

Main Methods:

• Utilized graph neural networks (GNNs) for feature extraction from molecular data.
• Employed transfer learning by pre-training a model on solubility prediction.
• Applied the enhanced GNN model to predict oral bioavailability.

Main Results:

• Achieved an average accuracy of 0.797 for oral bioavailability prediction.
• Obtained an F1 score of 0.840 and an AUC-ROC of 0.867.
• Outperformed previous studies using the same test dataset.

Conclusions:

• GNNs with transfer learning provide a powerful and efficient method for predicting oral bioavailability.
• Automated feature selection by GNNs reduces the need for domain expertise.
• The developed model shows significant potential for accelerating drug discovery pipelines.