A Multibranch Neural Network for Drug-Target Affinity Prediction Using Similarity Information

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Predicting drug-target affinity (DTA) is crucial for efficient drug discovery.
• Current graph-based deep learning models often analyze drugs and proteins in isolation, limiting interaction insights.
• Existing network-based models rely on knowledge graphs, which can be resource-intensive.

Purpose of the Study:

• To develop a novel deep learning framework for DTA prediction that incorporates drug and protein similarity information.
• To overcome the limitations of models that only use molecular structure or rely on external knowledge graphs.
• To improve the accuracy and efficiency of DTA prediction for accelerated drug discovery.

Main Methods:

• Proposed a novel network framework that autonomously extracts drug and protein similarity information.
• Designed a multibranch neural network, GASI-DTA, integrating similarity, sequence, and molecular structure data.
• Evaluated the model on two benchmark datasets and three cold-start scenarios.

Main Results:

• GASI-DTA significantly outperformed state-of-the-art graph structure-based methods across most metrics.
• The model demonstrated substantial advantages over existing network-based models in DTA prediction.
• Experimental results validated the effectiveness of integrating similarity information for improved DTA prediction.

Conclusions:

• The proposed GASI-DTA model offers a powerful new approach for drug-target affinity prediction.
• Autonomous extraction of similarity information enhances model performance and reduces reliance on external knowledge graphs.
• This work provides a valuable tool for accelerating the drug discovery pipeline.