Recent advancements in spatial transcriptomics data analysis have enabled gene expression studies at the single-cell level while preserving the spatial information of cells across tissues. Previously, in single-cell RNA sequencing, the spatial interrelationships between cells were lost. Cancer cell annotation is important to understand the underlying patterns in cancer research and the cellular heterogeneity of the tumor microenvironment. Recently, leveraging the learned knowledge from available data to make inferences for unequal data distributions has helped researchers build efficient computational models. In this work, we focus on a domain generalization model for malignant cell annotation in brain cancer spatial transcriptomics data by leveraging a two-module deep learning framework. We trained the deep learning model with new spatial data in the source and target domains of the domain generalization model. Moreover, previous approaches have not adopted brain cancer data as the source domain for model training. In our proposed approach, we trained the domain generalization model using a new source domain data and tested the model with other brain cancer spatial data. The overall results reflect the proposed approach’s performance in terms of accuracy on unseen test data in another domain.