Multiscale Motif-Aware Relation Graph Structure for Drug-Target Binding Affinity Prediction.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Drug-target binding affinity (DTA) prediction is critical for efficient drug discovery.
• Existing methods often rely on 1D SMILES, neglecting vital molecular structural details.
• Incorporating structural information is key to accurately predicting drug properties.

Purpose of the Study:

• To develop a novel descriptor, the multiscale motif-aware relation graph (MMRG), for improved DTA prediction.
• To leverage motif-level structural and topological information for enhanced drug representation.
• To outperform existing DTA prediction methods by incorporating advanced graph-based features.

Main Methods:

• Proposed a novel MMRG construction approach for drugs.
• Implemented multiscale motif-aware learning to extract structural information from various motif sizes.
• Utilized a relation graph construction to capture topological information from chemical bonds.
• Employed a graph convolutional network (GCN) for learning latent features from MMRGs.

Main Results:

• The proposed MMRG approach significantly improved DTA prediction accuracy on the Davis and KIBA datasets.
• Achieved an average performance improvement of 15.61% on the Davis dataset and 8.50% on the KIBA dataset compared to state-of-the-art methods.
• Demonstrated the effectiveness of MMRG in capturing essential structural and topological drug features.

Conclusions:

• MMRG descriptors provide a more comprehensive representation of drug molecules than SMILES.
• The proposed method offers a significant advancement in drug-target affinity prediction accuracy.
• This approach holds promise for accelerating the drug discovery pipeline by enabling more precise predictions.