MutualDTA: An Interpretable Drug-Target Affinity Prediction Model Leveraging Pretrained Models and Mutual Attention.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug-target affinity (DTA) prediction is crucial for accelerating drug development.
• Existing deep learning models for DTA prediction suffer from insufficient data representation, incomplete feature extraction, and lack of interpretability.
• Addressing these limitations is essential for advancing computational drug discovery.

Purpose of the Study:

• To propose MutualDTA, an interpretable deep learning model for accurate DTA prediction.
• To enhance data representation and feature extraction for improved DTA modeling.
• To provide interpretability in drug-target binding predictions.

Main Methods:

• Utilizing pretrained models for accurate drug and target representations.
• Employing specialized modules for comprehensive hidden feature extraction.
• Implementing a Mutual-Attention module for modeling intermolecular interactions and identifying binding sites.

Main Results:

• MutualDTA outperformed 12 state-of-the-art models on two benchmark datasets.
• Attention visualization demonstrated MutualDTA's ability to identify partial interaction sites, enhancing interpretability.
• Application of MutualDTA for screening Alzheimer's disease-related targets identified potential drug candidates.

Conclusions:

• MutualDTA offers a reliable and interpretable approach for DTA prediction.
• The model aids drug developers by reducing the search space for binding sites.
• MutualDTA shows promise in accelerating the identification of effective drug candidates for diseases like Alzheimer's.