MarsGT is a multi-omics analysis tool for rare population inference using a single-cell graph transformer. It identifies rare cell populations using a probability-based heterogeneous graph transformer on single-cell multi-omics data. MarsGT outperforms existing tools in identifying rare cells across 550 simulated and four real human datasets. In mouse retina data, it reveals unique subpopulations of rare bipolar cells and a Müller glia cell subpopulation. In human lymph node data, MarsGT detects an intermediate B cell population potentially acting as lymphoma precursors. In human melanoma data, it identifies a rare MAIT-like population impacted by a high IFN-I response and reveals the mechanism of immunotherapy. MarsGT offers biological insights and suggests potential strategies for early detection and therapeutic intervention of disease. Multicellular organisms encompass a diverse range of specialized cells. Identifying these cell types is pivotal in immunotherapy and clinical scenarios, as it illuminates immune mechanisms, aids in devising targeted therapies, and bolsters personalized medicine by unmasking the unique cellular makeup of each patient. However, difficulties surface when encountering rare or transiently expressed cells. Despite their scarcity, rare cell populations step up to play crucial roles in a variety of biological processes. A refined grasp of these rare cell populations, culminating in a more detailed depiction, will illuminate our understanding of tumor microenvironments and the intricate mechanisms that steer the responses to immunotherapy. The advent of single-cell RNA sequencing (scRNA-seq) has vastly improved our ability to identify individual cell types, offering high-resolution molecular profiles that illuminate cellular diversity and the complex dynamics of gene expression within specific cells. Most existing rare cell identification tools confront several challenges, such as high false positives when inferring rare populations, limited performance with complex samples like tumor biopsy. These issues could stem from the limited representation of rare cells, which may lead to inaccurate grouping with more prevalent cell populations when solely relying on gene expression data. This pursuit can be further accelerated by technological innovations like single-cell ATAC sequencing (scATAC-seq). When synergistically used with scRNA-seq, these methodologies provide partial regulatory data concerning enhancer regions pivotal in preserving cell type identities. This invaluable information can be tapped into for the construction of gene regulatory networks, thereby unraveling critical insights into the nature and function of rare cell populations. Meanwhile, graph neural networks have recently demonstrated profound proficiency in deciphering complex biological data, offering robust backing for the precise analysis and study of scMulti-omics data. The implementation of the heterogeneous graph transformer provides a unified framework that amalgamates diverse single-cell data types, thereby facilitating a comprehensive understanding of cellular heterogeneity. This approach unveils the intricate interplay among various cell types within complex cellular landscapes, enhancing our comprehension of biological systems and bolstering opportunities for precision therapeutic interventions. To fill the gap and validate the theory, we developed MarsGT (Multi-omics analysis for rare population inference using single-cell Graph Transformer), an end-toMarsGT is a multi-omics analysis tool for rare population inference using a single-cell graph transformer. It identifies rare cell populations using a probability-based heterogeneous graph transformer on single-cell multi-omics data. MarsGT outperforms existing tools in identifying rare cells across 550 simulated and four real human datasets. In mouse retina data, it reveals unique subpopulations of rare bipolar cells and a Müller glia cell subpopulation. In human lymph node data, MarsGT detects an intermediate B cell population potentially acting as lymphoma precursors. In human melanoma data, it identifies a rare MAIT-like population impacted by a high IFN-I response and reveals the mechanism of immunotherapy. MarsGT offers biological insights and suggests potential strategies for early detection and therapeutic intervention of disease. Multicellular organisms encompass a diverse range of specialized cells. Identifying these cell types is pivotal in immunotherapy and clinical scenarios, as it illuminates immune mechanisms, aids in devising targeted therapies, and bolsters personalized medicine by unmasking the unique cellular makeup of each patient. However, difficulties surface when encountering rare or transiently expressed cells. Despite their scarcity, rare cell populations step up to play crucial roles in a variety of biological processes. A refined grasp of these rare cell populations, culminating in a more detailed depiction, will illuminate our understanding of tumor microenvironments and the intricate mechanisms that steer the responses to immunotherapy. The advent of single-cell RNA sequencing (scRNA-seq) has vastly improved our ability to identify individual cell types, offering high-resolution molecular profiles that illuminate cellular diversity and the complex dynamics of gene expression within specific cells. Most existing rare cell identification tools confront several challenges, such as high false positives when inferring rare populations, limited performance with complex samples like tumor biopsy. These issues could stem from the limited representation of rare cells, which may lead to inaccurate grouping with more prevalent cell populations when solely relying on gene expression data. This pursuit can be further accelerated by technological innovations like single-cell ATAC sequencing (scATAC-seq). When synergistically used with scRNA-seq, these methodologies provide partial regulatory data concerning enhancer regions pivotal in preserving cell type identities. This invaluable information can be tapped into for the construction of gene regulatory networks, thereby unraveling critical insights into the nature and function of rare cell populations. Meanwhile, graph neural networks have recently demonstrated profound proficiency in deciphering complex biological data, offering robust backing for the precise analysis and study of scMulti-omics data. The implementation of the heterogeneous graph transformer provides a unified framework that amalgamates diverse single-cell data types, thereby facilitating a comprehensive understanding of cellular heterogeneity. This approach unveils the intricate interplay among various cell types within complex cellular landscapes, enhancing our comprehension of biological systems and bolstering opportunities for precision therapeutic interventions. To fill the gap and validate the theory, we developed MarsGT (Multi-omics analysis for rare population inference using single-cell Graph Transformer), an end-to