LKE-DTA: predicting drug-target binding affinity with large language model representations and knowledge graph...

Area of Science:

• Computational chemistry
• Bioinformatics
• Artificial intelligence in drug discovery

Background:

• Accurate drug-target binding affinity (DTA) prediction is crucial for efficient drug discovery.
• Existing computational methods face challenges in integrating diverse biomedical data and modeling complex molecular interactions.

Purpose of the Study:

• To develop a novel deep learning framework, LKE-DTA, for enhanced DTA prediction.
• To leverage large language models (LLMs) and knowledge graphs (KGs) for comprehensive molecular representations.
• To improve the accuracy and generalization of DTA prediction models.

Main Methods:

• Developed LKE-DTA, a deep learning framework integrating LLMs and KGs.
• Employed a Dual Multi-Head Attention mechanism for dynamic fusion of heterogeneous embeddings.
• Conducted comprehensive evaluations on benchmark datasets (Davis, KIBA) using fivefold cross-validation and cold-start scenarios.

Main Results:

• LKE-DTA consistently outperformed state-of-the-art methods across benchmark datasets.
• Achieved significant reductions in Mean Squared Error (MSE) and Mean Absolute Error (MAE), and improvements in Concordance Index (CI) and Pearson correlation coefficient (r).
• Demonstrated strong generalization in cold-start settings (cold drug and cold target) and on an independent test set.

Conclusions:

• The integration of LLMs and KGs offers a powerful approach for DTA prediction.
• LKE-DTA provides a robust and accurate framework for advancing drug design and precision medicine.
• This work highlights the potential of hybrid AI models in addressing complex biomedical challenges.