DeepExpDR: Drug Response Prediction through Molecular Topological Grouping and Substructure-Aware Expert.

Area of Science:

• Computational Biology
• Drug Discovery
• Artificial Intelligence in Oncology

Background:

• Tumor heterogeneity complicates cancer treatment and drug response prediction.
• Current experimental methods for drug response verification are time-consuming and costly.
• Existing deep learning models often overlook molecular topological properties in drug response prediction.

Purpose of the Study:

• To develop DeepExpDR, a deep expert framework for accurate drug response prediction.
• To incorporate molecular topological features into drug response prediction models.
• To improve the efficiency and effectiveness of anticancer drug development and precision medicine.

Main Methods:

• Pretraining a self-supervised clustering model to group drugs by molecular scaffold similarity.
• Assigning drug groups to specialized substructure-aware experts.
• Utilizing substructure sensing networks integrating molecular topology, gene expression, and drug response correlation matrices to predict IC50 values.

Main Results:

• DeepExpDR achieved state-of-the-art performance in both warm and cold settings for regression and classification tasks.
• The framework demonstrated effectiveness in predicting unknown cancer drug responses through a case study.
• Experimental results validate the model's ability to leverage molecular substructure information.

Conclusions:

• DeepExpDR offers a powerful computational approach for drug response prediction, outperforming existing methods.
• The framework's ability to consider molecular topology enhances drug feature extraction and prediction accuracy.
• DeepExpDR facilitates advancements in precision medicine and accelerates anticancer drug discovery.