ImageDTA: A Simple Model for Drug-Target Binding Affinity Prediction.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Bioinformatics

Background:

• Predicting drug-target binding affinity (DTA) is essential for efficient drug discovery.
• Current artificial intelligence (AI) methods, including deep neural networks, require improvements in accuracy, complexity, and efficiency.
• Convolutional Neural Networks (CNNs) demonstrate effective learning capabilities, even with limited data.

Purpose of the Study:

• To introduce ImageDTA, a novel prediction method based on a multiscale 2-dimensional convolutional neural network (CNN).
• To leverage CNNs' image processing capabilities for enhanced DTA prediction.
• To improve upon the accuracy, training, and inference efficiency of existing DTA prediction models.

Main Methods:

• ImageDTA utilizes a multiscale 2-dimensional CNN architecture.
• Molecular data is encoded using Simplified Molecular Input Line Entry System (SMILES) strings and treated as "images" for CNN processing.
• Visualization techniques are employed to optimize convolutional kernel sizes for interpretability.

Main Results:

• ImageDTA demonstrates higher training and inference efficiency compared to pretrained large models.
• ImageDTA achieves superior accuracy and interpretability over attention-based graph neural network models.
• The use of CNNs with image-like molecular representations enhances model performance.

Conclusions:

• ImageDTA offers a promising AI-driven approach for accurate and efficient drug-target binding affinity prediction.
• The method's unique image-based processing of molecular data enhances CNN learning.
• ImageDTA provides improved interpretability, aiding in the understanding of drug-target interactions.