A Length-Adaptive LLM-Enhanced Method for Drug-Target Interaction Prediction

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for drug repurposing and discovery.
• Existing DTI methods often rely on structural information, limiting their ability to capture complex biomedical semantics.
• Insufficient semantic information and feature imbalance due to varying protein lengths hinder the performance of current DTI prediction models.

Purpose of the Study:

• To propose BioLA-DTI, a novel method for drug-target interaction prediction enhanced by large language models (LLMs).
• To address the limitations of existing methods by incorporating rich semantic information and a length-adaptive feature fusion strategy.
• To improve the accuracy and robustness of DTI prediction, particularly in challenging cold-start scenarios.

Main Methods:

• Leveraging LLMs to extract comprehensive semantic features for both drugs and proteins.
• Developing a length-adaptive fusion module to dynamically integrate LLM-derived semantic features with structural features based on protein sequence length.
• Implementing a novel weighting strategy to mitigate feature imbalance caused by variations in protein lengths.

Main Results:

• BioLA-DTI significantly outperforms existing baseline methods on two benchmark datasets across multiple evaluation metrics.
• The proposed method demonstrates superior performance in cold-start scenarios, indicating its effectiveness in predicting interactions for novel drug-target pairs.
• Ablation studies confirm the substantial contributions of both LLM enhancement and the length-adaptive fusion module to the overall performance.

Conclusions:

• BioLA-DTI represents a significant advancement in drug-target interaction prediction by effectively integrating LLM-based semantic information and addressing protein length variation.
• The model's enhanced predictive power, especially in cold-start settings, offers valuable insights for accelerating drug discovery and repurposing efforts.
• The developed length-adaptive approach provides a robust solution for handling feature imbalance in biological sequence data, applicable beyond DTI prediction.