DEAttentionDTA: protein-ligand binding affinity prediction based on dynamic embedding and self-attention.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Predicting protein-ligand binding affinity is essential for drug development.
• Existing methods often require complex 3D protein structures, which are difficult to obtain.
• Integrating protein and ligand sequence information and identifying active sites remain challenges.

Purpose of the Study:

• To develop an accurate and efficient model for predicting protein-ligand binding affinity using only sequence data.
• To address the limitations of existing methods that rely on 3D structural information.

Main Methods:

• Proposed DEAttentionDTA, a neural network model utilizing dynamic word embeddings and a self-attention mechanism.
• Input includes 1D amino acid sequences (global and active site features) and ligand SMILES strings.
• A 1D convolutional neural network encodes sequences, and a self-attention mechanism correlates them.

Main Results:

• DEAttentionDTA achieved superior performance compared to mainstream tools on the same dataset.
• The model demonstrated effectiveness in predicting binding affinity for the p38 protein family.
• The approach successfully utilizes 1D sequence information, bypassing the need for 3D structures.

Conclusions:

• DEAttentionDTA offers a promising deep learning approach for predicting protein-ligand binding affinity.
• The model's reliance on sequence data makes it broadly applicable in drug discovery.
• The open-source availability facilitates further research and development.