AttentionDTA: Drug-Target Binding Affinity Prediction by Sequence-Based Deep Learning With Attention Mechanism.

Area of Science:

• Computational Biology
• Drug Discovery
• Machine Learning

Background:

• Drug-target relations (DTRs) are crucial for drug development, but traditional methods often treat them as binary interactions (DTIs), lacking quantitative affinity and dose-dependence information.
• Existing deep learning models for drug-target affinity (DTA) prediction, while effective, suffer from a lack of biological interpretability due to their black-box nature.
• The increasing availability of drug-protein binding affinity data presents an opportunity to reframe DTR prediction as a regression problem, yielding more detailed insights.

Purpose of the Study:

• To propose AttentionDTA, a novel deep learning model designed to predict drug-target affinities (DTAs) with enhanced biological interpretability.
• To leverage attention mechanisms to identify and focus on critical subsequences within drug (SMILES) and protein sequences that influence binding affinity.
• To outperform existing state-of-the-art deep learning methods in DTA prediction across multiple benchmark datasets.

Main Methods:

• Utilized two separate 1D Convolutional Neural Networks (1D-CNNs) to extract semantic features from drug SMILES strings and protein amino acid sequences.
• Developed and integrated a two-side multi-head attention mechanism to explore and model the intricate relationships between drug and protein features.
• Evaluated AttentionDTA on three established DTA benchmark datasets: Davis, Metz, and KIBA, and additionally tested on an IC50 dataset for binding site localization.

Main Results:

• AttentionDTA demonstrated superior performance compared to state-of-the-art deep learning methods across various evaluation metrics on the Davis, Metz, and KIBA datasets.
• The attention-based approach effectively captured relevant drug and protein features, leading to more accurate DTA predictions.
• Visualization of attention weights confirmed the model's ability to identify biologically significant binding sites and subsequences.

Conclusions:

• AttentionDTA offers a powerful and interpretable deep learning framework for predicting drug-target affinities, advancing drug discovery efforts.
• The attention mechanism is key to extracting salient features from sequential data, improving the accuracy and interpretability of DTA prediction models.
• The model's ability to highlight important sequence regions provides valuable insights into drug-target interactions, facilitating rational drug design.