Multitask Learning on Graph Convolutional Residual Neural Networks for Screening of Multitarget Anticancer Compounds

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Experimental screening for anticancer drugs is time-consuming and resource-intensive.
• Computational methods are needed to accelerate the identification of potential anticancer compounds.
• Multitarget drug discovery aims to identify compounds effective against multiple cancer types.

Purpose of the Study:

• To develop an advanced computational model for identifying multitarget anticancer compounds.
• To improve the efficiency and accuracy of anticancer drug candidate screening.
• To provide a publicly accessible tool for researchers.

Main Methods:

• Utilized multitask learning on graph convolutional residual neural networks (GCRNs) with two shortcut connection types.
• Trained and evaluated the iACP-GCR model on the NCI-60 dataset, encompassing nine cancer types.
• Compared the model's performance against three advanced computational multitask learning methods.

Main Results:

• The iACP-GCR model demonstrated superior performance compared to existing advanced computational methods.
• The integration of dual shortcut connections enhanced prediction efficiency.
• The model successfully identified multitarget anticancer compounds with high accuracy.

Conclusions:

• iACP-GCR offers a powerful and efficient computational approach for anticancer drug discovery.
• The model's effectiveness in identifying multitarget compounds holds significant promise for cancer therapy.
• A public web server has been deployed to facilitate broader research community access and utilization.