Single-cell transcriptomics and proteomics data represent an incredibly rich source of information about patient states over the course of disease and treatment. However, the noise and complexity inherent in this data have often prevented researchers from utilizing its full potential, especially when comparing multiple samples between conditions. Scailyte's ScaiVision platform offers a tailor-made solution, using a representation learning approach to extract specific signals from single-cell data that are most relevant to predicting clinical endpoints. We will demonstrate ScaiVision's power in an application of biomarker discovery from CAR-T cells.