AACDR: Integrating Graph Isomorphism Networks and Asymmetric Adversarial Domain Adaptation for Cancer Drug Response...

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• Predicting cancer drug response is crucial for personalized medicine.
• Preclinical data often has distribution discrepancies compared to clinical data, limiting prediction accuracy.
• Existing methods overlook these distributional differences.

Purpose of the Study:

• To develop a novel cancer drug response prediction method addressing distribution discrepancies.
• To enhance the accuracy and stability of predictions for clinical applications.
• To facilitate personalized cancer treatment strategies.

Main Methods:

• Introduced asymmetric adversarial domain adaptation (AACDR) for knowledge transfer.
• Utilized graph isomorphism networks for enhanced drug feature extraction.
• Validated the model on cancer patient datasets, patient-derived xenograft (PDX), and copy number variation (CNV) data.

Main Results:

• AACDR demonstrated effective knowledge transfer from cell lines to patients.
• The method outperformed existing approaches in cancer drug response prediction.
• AACDR showed generalizability across various distributional discrepancies and cancer representation methods.

Conclusions:

• AACDR accurately predicts therapeutic outcomes for clinical cancer records.
• The model can recommend personalized therapeutic options for cancer patients.
• This approach holds significant practical importance for advancing personalized cancer therapy.