A Multi-modal Drug Target Affinity Prediction Based on Graph Features and Pre-trained Sequence Embeddings.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target affinity (DTA) prediction is crucial for efficient drug discovery, reducing experimental costs.
• Existing deep learning methods often rely on single data modalities (either drug or target features).

Purpose of the Study:

• To propose MGSDTA, a multi-modal deep learning framework for enhanced DTA prediction.
• To integrate diverse molecular representations for improved computational drug discovery.

Main Methods:

• Feature extraction from drug and target molecular graphs.
• Utilizing advanced self-supervised models (Mol2vec, ProtVec) for continuous sequence embeddings.
• Employing a weighted fusion module to combine multi-modal features for DTA prediction.

Main Results:

• MGSDTA outperforms existing single-modal DTA prediction methods.
• The integration of graph and sequence features significantly enhances prediction performance.
• Validation on benchmark datasets confirms the efficacy of the multi-modal approach.

Conclusions:

• MGSDTA offers a more accurate and efficient computational method for DTA prediction.
• Multi-modal data integration is a promising strategy for advancing drug discovery.
• The proposed method can accelerate the screening of potential drug candidates.