DTA-GTOmega: Enhancing Drug-Target Binding Affinity Prediction with Graph Transformers Using OmegaFold Protein...

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Drug-protein interactions are vital for drug development but challenging to model accurately.
• Existing methods struggle with representing 3D protein structures and complex drug-target relationships.

Purpose of the Study:

• To develop a new computational method, DTA-GTOmega, for predicting drug-target binding affinity.
• To improve the accuracy and generalization capabilities of drug-target interaction (DTI) predictions.

Main Methods:

• Utilized OmegaFold for 3D protein structure prediction and target graph construction.
• Processed drug SMILES sequences with RDKit to generate drug graphs.
• Employed multi-layer graph transformer and co-attention modules to integrate drug and target features.

Main Results:

• DTA-GTOmega significantly improved binding affinity prediction accuracy.
• Outperformed existing methods on KIBA, Davis, and BindingDB_Kd datasets under a cold-start setting.
• Demonstrated competitive performance on real-world DrugBank data and disease-specific DTI scenarios.

Conclusions:

• DTA-GTOmega offers a robust and accurate approach for predicting drug-target binding affinity.
• The method shows strong generalization capabilities and potential for handling imbalanced DTI data.
• This advancement aids in elucidating drug mechanisms and optimizing drug development.