MSCMLCIDTI: Drug-Target Interaction Prediction Based on Multiscale Feature Extraction and Deep Interactive Attention...

Area of Science:

• Computational Biology
• Drug Discovery
• Artificial Intelligence

Background:

• Drug-target interaction (DTI) prediction is vital for accelerating drug discovery.
• Current deep learning methods struggle with local feature representation and modeling interactions between drug and target data.

Purpose of the Study:

• To introduce a novel deep learning framework, the multiscale feature extraction coupled multilayer cross-interaction network (MSCMLCIDTI), to enhance DTI prediction.
• To address limitations in existing models regarding feature representation and interaction modeling.

Main Methods:

• Utilized multiscale convolutional blocks for extracting structural fingerprints of drugs and amino acid sequences at various scales.
• Employed gated attention mechanisms to derive multidimensional features for identifying critical binding sites.
• Implemented a deep cross-interaction mechanism with multilayer attention to model complex relationships between drug substructures and protein fragments.

Main Results:

• The MSCMLCIDTI framework demonstrated superior predictive accuracy compared to existing state-of-the-art models.
• Validated performance across four open-access benchmark datasets, confirming enhanced capability in identifying interaction signatures.
• Improved representation of local features and interaction modeling between drugs and targets.

Conclusions:

• The proposed MSCMLCIDTI model offers a significant advancement in drug-target interaction prediction.
• This framework has the potential to accelerate the drug discovery pipeline by improving prediction accuracy.
• The multiscale feature extraction and cross-interaction mechanisms are key to the model's enhanced performance.