IPM-DTI: An Interaction-Pattern-Driven Multimodal Framework for Drug-Target Interaction Prediction via Knowledge...

Area of Science:

• Computational drug discovery
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Accurate drug-target interaction (DTI) prediction is vital for computational drug discovery.
• Existing multimodal methods often overlook complex, high-order interaction patterns.
• Knowledge graphs (KGs) can reveal indirect associations but face computational challenges in large search spaces.

Purpose of the Study:

• To develop a novel multimodal DTI prediction framework, IPM-DTI, that leverages learned interaction patterns.
• To integrate heterogeneous data including protein sequences, molecular graphs, and KG structures for enhanced DTI prediction.
• To address the computational challenges of KG traversal for identifying latent drug-target associations.

Main Methods:

• Constructed an explicit DTI knowledge graph (KG) from DTI databases.
• Employed multimodal encoders to map drugs and targets into a unified latent space, extracting cross-modal features.
• Decomposed the global DTI KG into semantic subspaces and utilized an RL-based reasoning module for navigating these subspaces.
• Integrated structure- and RL-derived results via a multisource fusion module for comprehensive DTI prediction.

Main Results:

• The proposed IPM-DTI framework demonstrated superior performance compared to single-modal and single-embedding baseline methods.
• The integration of heterogeneous data and learned interaction patterns significantly improved DTI prediction accuracy.
• The RL-based reasoning module effectively navigated KG subspaces to identify latent drug-target associations.

Conclusions:

• The IPM-DTI framework effectively captures complex, high-order interaction patterns for accurate DTI prediction.
• Integrating multimodal data with KG structures and RL-guided reasoning enhances computational drug discovery.
• This approach offers a promising direction for improving the efficiency and accuracy of identifying potential drug candidates.