Document resource
Background Traditional gating alone is insufficient for the comprehensive analysis of high-dimensional cytometry data. As the number of markers increases, the number of possible cell phenotypes grows exponentially, making manual identification of novel populations impractical. For example, a 42-marker panel generates 861 potential biaxial plots, requiring over 240 hours to manually review all possible plots. Unsupervised computational methods provide a powerful alternative, enabling an unbiased and comprehensive survey of the entire immune landscape. By using modern algorithms for clustering and UMAP for visualization, researchers can identify clinically significant cell populations that would be difficult for a human to find by eye. Applying these methods is particularly useful when identifying immunological shifts that precede immune-related adverse events (IRAEs) in patients treated with checkpoint inhibitors.Methods We analyzed peripheral blood mononuclear cell (PBMC) samples from 29 melanoma patients (70 samples total) treated with anti-PD-1 therapy at the Huntsman Cancer Institute. Using a 43-marker mass cytometry panel, clinicians collected samples at up to three relevant timepoints: baseline, pre-IRAE, and maximal IRAE severity. We applied our new unsupervised analysis pipeline to this high-dimensional data to identify immune cell populations and track their changes associated with IRAE development.Results Our unbiased, data-driven approach successfully characterized the complex immune profiles of all 70 samples. The analysis identified a unique immune cell population that was not previously identified and is relevant to IRAE. This population, characterized by a distinct expression profile that defies canonical lineage definitions, would likely be missed by a conventional gating strategy focused on known cell types. The discovery of this novel cell state demonstrates the capability of our analysis pipeline to resolve previously uncharacterized populations that may be directly involved in the development of IRAEs.Conclusions The application of our unsupervised learning workflow to this longitudinal melanoma cohort provides a framework for uncovering signatures of immune-related toxicity. This approach not only overcomes the limitations of manual gating but also offers a method for discovering novel biomarkers in complex clinical datasets. The future identification of unique cell populations associated with IRAEs can lead to a better understanding of toxicity mechanisms and may inform predictive strategies for patients undergoing immunotherapy.Acknowledgements We are grateful for the support from Dr. Siwen Hu-Lieskovan and colleagues at the Hunstman Cancer Institute for providing the PBMC samples analyzed in this study.