Decoupled dual-granularity rebalanced pyramid network for drug-target interaction prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery but faces challenges due to size differences between drugs and proteins, hindering accurate binding site prediction.
• Modality imbalance in learning biases can undermine the effectiveness of multimodal representations in DTI prediction.

Purpose of the Study:

• To propose DDGR-DTI, a novel framework addressing challenges in DTI prediction, specifically binding site identification and modality imbalance.
• To enhance the accuracy and generalization ability of DTI prediction models.

Main Methods:

• The proposed DDGR-DTI utilizes a Decoupled Dual-Granularity Framework, dividing the DTI task into macro (modality-based subtasks) and micro (within-subtask representation) levels.
• A dual-stream attention module is employed for fine-grained substructure-level interactions to accurately identify binding sites.
• A Rebalanced Pyramid Network (RPN) is used to mitigate modality imbalance in multimodal fusion via hierarchical aggregation.

Main Results:

• DDGR-DTI demonstrated superior performance compared to existing state-of-the-art models in benchmark tests.
• The model showed enhanced prediction performance and generalization ability in drug-target interaction prediction.
• The framework effectively addressed challenges related to binding site identification and modality imbalance.

Conclusions:

• DDGR-DTI offers a significant advancement in drug-target interaction prediction by effectively handling size differences and modality imbalance.
• The proposed framework improves the accuracy and reliability of DTI prediction, facilitating more efficient drug discovery pipelines.
• The source code and dataset are publicly available, promoting further research and development in the field.