GFLearn: Generalized Feature Learning for Drug-Target Binding Affinity Prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence

Background:

• Accurate prediction of drug-target binding affinity is crucial for efficient drug discovery and development.
• Current deep learning models often struggle with predicting affinities for new drugs or targets due to data dependency.
• Performance degradation is common when models encounter data distribution shifts.

Purpose of the Study:

• To develop a novel Generalized Feature Learning (GFLearn) model for robust drug-target binding affinity prediction.
• To improve the generalizability of predictive models for unseen drugs and targets.
• To mitigate performance degradation caused by data distribution shifts in drug discovery.

Main Methods:

• Integration of Graph Neural Networks (GNNs) with a self-supervised invariant feature learning module.
• Extraction of robust and generalizable features from both drug and target molecules.
• Extensive experimental validation on diverse datasets across new drug, new target, and combined scenarios.

Main Results:

• The GFLearn model consistently outperformed state-of-the-art methods in predicting binding affinities for new drugs and targets.
• Demonstrated robustness across various prediction tasks and validated generalizability through cross-dataset evaluations.
• Case studies confirmed accurate predictions for specific drug-target pairs, aiding drug screening and repurposing.

Conclusions:

• The GFLearn model offers a significant advancement in predicting drug-target binding affinity, particularly for novel entities.
• Its ability to handle data distribution shifts enhances reliability in real-world drug discovery applications.
• GFLearn provides valuable insights for accelerating drug screening and repurposing efforts.