Anticancer drug response prediction integrating multi-omics pathway-based difference features and multiple deep...

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Individualized prediction of cancer drug sensitivity is crucial for precision medicine.
• Current predictive methods face challenges in accurately forecasting patient responses and understanding inter-individual variations.
• Omics data and drug chemical structures offer rich information for predictive modeling.

Purpose of the Study:

• To develop a deep learning model (PASO) for accurate prediction of anticancer drug sensitivity in cell lines.
• To integrate diverse omics data with drug molecular structures for enhanced predictive power.
• To identify key biological pathways and drug features influencing drug response.

Main Methods:

• Utilized omics data (gene expression, mutation, copy number variations) and SMILES representations of drugs as input.
• Developed a deep learning architecture integrating multi-scale convolutional networks, transformer encoders, and attention mechanisms.
• Employed statistical methods to derive pathway-based cell line features.

Main Results:

• The PASO model demonstrated superior accuracy in predicting anticancer drug sensitivity compared to existing methods.
• Identified specific drug classes (PARP and Topoisomerase I inhibitors) highly sensitive to small cell lung cancer (SCLC).
• The model successfully highlighted relevant biological pathways and critical drug structural components.

Conclusions:

• The PASO model shows significant potential as a tool to support individualized cancer treatment strategies.
• The approach provides a robust framework for integrating multi-modal data in drug sensitivity prediction.
• The developed methods are publicly available to facilitate further research and clinical application.