MML-DTI: Multimanifold Learning with Hyperbolic Graph Neural Networks for Enhanced Drug-Target Interaction...

Area of Science:

• Computational biology
• Drug discovery
• Machine learning

Background:

• Accurate drug-target interaction (DTI) prediction is vital for drug discovery and repositioning.
• Existing deep learning models often struggle with the hierarchical nature of biological data due to their Euclidean space design.
• Hyperbolic space offers a promising alternative for representing complex hierarchical relationships.

Purpose of the Study:

• To develop a novel multimanifold learning framework for enhanced DTI prediction.
• To effectively integrate multimodal features from drugs and targets.
• To leverage hyperbolic geometry for superior representation of biological data.

Main Methods:

• Proposed a multimanifold learning framework integrating hyperbolic and Euclidean spaces.
• Employed a Hyperbolic Graph Neural Network (HGNN) for extracting hierarchical features from molecular graphs.
• Utilized a Multi-Manifold Feature Fusion Module to combine HGNN features, chemical fingerprints, and language model embeddings.

Main Results:

• The proposed framework achieved superior performance compared to state-of-the-art Euclidean-based DTI prediction methods.
• Demonstrated the advantages of hyperbolic geometry in capturing hierarchical features from non-Euclidean biological data.
• Validated the effectiveness of multimanifold feature fusion for DTI prediction.

Conclusions:

• Hyperbolic geometry provides significant benefits for DTI prediction by effectively modeling hierarchical data structures.
• The multimanifold feature fusion approach shows considerable potential for advancing DTI prediction.
• This framework offers a promising direction for future drug discovery and repositioning efforts.