scXDR: drug response prediction across single-cell datasets via heterogeneous network transfer learning

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Single-cell drug response prediction is crucial for personalized medicine.
• Current methods integrating bulk and single-cell RNA sequencing data have limitations due to data discrepancies and assumptions.
• Accurate cell-level drug response prediction is needed for improved treatment strategies.

Purpose of the Study:

• To develop a novel heterogeneous network transfer learning model, scXDR, for accurate drug response prediction at the single-cell level.
• To overcome the limitations of existing methods that transfer bulk data labels to single cells.
• To provide a robust framework for predicting drug efficacy across diverse single-cell datasets.

Main Methods:

• Proposed a heterogeneous network transfer learning model (scXDR).
• Integrated features and associations among drugs, genes, and cells using heterogeneous networks.
• Employed message passing, feature and structure alignment, and structure reconstruction for drug-cell score prediction.

Main Results:

• scXDR significantly outperformed existing transfer learning methods in various scenarios.
• Achieved excellent prediction performance at both individual cell and cell group levels.
• Demonstrated the importance of individual model components through analysis.

Conclusions:

• scXDR offers a superior approach for single-cell drug response prediction compared to existing methods.
• The model's performance in case studies highlights its potential for clinical applications, including drug screening and combination therapy.
• This research provides a valuable reference for advancing cell-level precision treatment strategies.