Single Cell Inference of Cancer Drug Response Using Pathway-Based Transformer Network

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Accurate prediction of cancer drug responses is vital for personalized medicine.
• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity and rare resistant populations.
• Distinct data distributions between bulk and scRNA-seq limit knowledge transfer from large cell line datasets.

Purpose of the Study:

• To develop a novel deep learning model, scPDS, for predicting drug sensitivities from scRNA-seq data.
• To overcome data distribution discrepancies between bulk and scRNA-seq for improved drug response prediction.
• To enhance the accuracy and efficiency of scRNA-seq analysis in cancer drug response prediction.

Main Methods:

• Developed a Transformer-based deep learning model, single-cell Pathway Drug Sensitivity (scPDS).
• Employed pathway activation transformation to predict drug sensitivities from scRNA-seq data.
• Integrated bulk RNA-seq data from extensive cell line datasets to improve model performance.

Main Results:

• scPDS demonstrated superior accuracy and computational efficiency compared to state-of-the-art methods.
• Analysis of breast cancer cells treated with bortezomib revealed dynamic changes in drug resistance.
• Identified drug-sensitive populations within resistant cells and predicted efficacy of combination therapies.
• Successfully distinguished between sensitive and resistant patients, correlating with survival outcomes.

Conclusions:

• scPDS provides a robust computational tool for predicting cellular drug responses from scRNA-seq data.
• The model offers valuable insights for optimizing personalized cancer treatment strategies.
• scPDS facilitates the identification of effective drug combinations and patient stratification for improved therapeutic outcomes.