DGDTA

DGDTA predicts drug–target binding affinity by integrating a dynamic graph attention network to model drugs from SMILES as molecular graphs and a bidirectional long short-term memory (Bi-LSTM) network to encode protein amino acid sequences for downstream affinity prediction in drug discovery and repositioning.

Key Features:

• Dynamic Graph Attention Network: Models drugs as graphs with atoms as nodes and bonds/interactions as edges and assigns dynamic attention scores to prioritize influential atoms and edges for affinity prediction.
• Bidirectional Long Short-Term Memory (Bi-LSTM): Encodes protein amino acid sequences to capture forward and backward sequential dependencies for protein feature extraction.
• Integration of Drug and Protein Features: Processes drug SMILES and protein sequences to produce feature vectors that represent molecular structure and protein characteristics.
• Prediction via Fully Connected Layer: Combines drug and protein feature vectors in a fully connected layer to generate drug–target binding affinity predictions.

Scientific Applications:

• Drug discovery: Predicts binding affinity to assist identification of potential therapeutic compounds.
• Drug repositioning: Supports evaluation of existing compounds against new targets by estimating affinity.
• Molecular interaction analysis: Facilitates analysis of atom- and residue-level contributions to drug–target binding through attention-weighted features.

Methodology:

Drugs are represented as graphs derived from SMILES and proteins as amino acid sequences; dynamic graph attention extracts drug molecular features, Bi-LSTM extracts protein sequential features, and the resulting vectors are concatenated and passed through a fully connected layer to predict binding affinity.