Drug resistance remains a primary cause of cancer treatment failure, while single-cell-level heterogeneity within the tumor microenvironment poses significant challenges for predicting drug sensitivity. This study proposes a framework for single-cell drug sensitivity prediction based on adversarial transfer learning. Our approach addresses the scarcity of annotated single-cell data by transferring drug response knowledge from bulk cell-line data (source domain) to single-cell data (target domain) through adversarial domain adaptation. We first establish a drug response classification model via supervised learning in the source domain, followed by adversarial alignment of feature distributions between domains. The experimental results demonstrate that our framework surpasses existing models in cross-domain transfer learning tasks, achieving 94% classification accuracy.