A drug response prediction method for single-cell tumors combining attention networks and transfer learning.

Introduction: Accurately predicting tumor cell line responses to therapeutic drugs is essential for personalized cancer treatment. Current methods using bulk cell data fail to fully capture tumor heterogeneity and the complex mechanisms underlying treatment responses.

Methods: This study introduces a novel method, ATSDP-NET (Attention-based Transfer Learning for Enhanced Single-cell Drug Response Prediction), which combines bulk and single-cell data. The model utilizes transfer learning and attention networks to predict drug responses in single-cell tumor data, after pre-training on bulk cell gene expression data. A multi-head attention mechanism is incorporated to enhance the model's expressive power and prediction accuracy by identifying gene expression patterns linked to drug reactions.

Results: ATSDP-NET outperforms existing methods in drug response prediction, as demonstrated on four single-cell RNA sequencing datasets. The model showed superior performance across multiple metrics, including recall, ROC, and average precision (AP). It accurately predicted the sensitivity and resistance of mouse acute myeloid leukemia cells to I-BET-762 and the sensitivity and resistance of human oral squamous cell carcinoma cells to cisplatin. Correlation analysis revealed a high correlation between predicted sensitivity gene scores and actual values (R = 0.888, < 0.001), while resistance gene scores also showed a significant correlation (R = 0.788, < 0.001). The dynamic process of cells transitioning from sensitive to resistant states was visualized using uniform manifold approximation and projection (UMAP).

Discussion: ATSDP-NET identifies critical genes linked to drug responses, confirming its predictions through differential gene expression scores and gene expression patterns. This method provides valuable insights into the mechanisms of drug resistance and offers potential for developing personalized treatment strategies.