MMSG-DTA: A Multimodal, Multiscale Model Based on Sequence and Graph Modalities for Drug-Target Affinity Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-Target Affinity (DTA) prediction is crucial for drug discovery but current methods struggle with molecular graph and protein feature extraction.
• Existing models often fail to capture global molecular graph features and rely on limited 1D protein sequences, hindering accurate interaction modeling.

Purpose of the Study:

• To develop an advanced model, MMSG-DTA, for improved Drug-Target Affinity prediction.
• To address limitations in capturing global molecular graph features and enhance protein representation for better drug-target interaction analysis.

Main Methods:

• Proposed a multimodal, multiscale model (MMSG-DTA) combining graph neural networks and Transformers for molecular graph feature extraction.
• Utilized a graph-based modality for improved protein feature extraction from amino acid sequences.
• Incorporated an attention-based feature fusion module to integrate diverse feature types for enhanced representation capacity.

Main Results:

• MMSG-DTA demonstrated superior performance in DTA prediction across three benchmark datasets (Davis, KIBA, Metz).
• The model effectively captured both local and global features from molecular graphs and improved protein representation.
• Experimental results showed MMSG-DTA outperformed several state-of-the-art DTA prediction methods.

Conclusions:

• The proposed MMSG-DTA model significantly advances the accuracy and robustness of drug-target interaction prediction.
• The multimodal and multiscale approach effectively integrates diverse data modalities for comprehensive feature learning.
• This work provides a powerful tool for accelerating drug discovery and development through enhanced DTA prediction.