Dimensionality reduction enables MRD visualization in single-sample flow cytometric analysis in acute leukemia

Minimal/measurable residual disease (MRD) analysis using multiparametric flow cytometry (MFC-MRD) is essential for therapy stratification in acute leukemia; however, conventional analysis is limited by analyst-dependent variability and complex manual gating. Although machine learning approaches have been proposed, their clinical implementation requires large reference datasets and standardized antibody panels. Therefore, we evaluated the feasibility of unsupervised dimensionality reduction as a practical reference-free strategy for single-sample MFC-MRD analysis. We retrospectively analyzed 58 bone marrow specimens from 32 patients with acute myeloid leukemia or B-lineage acute lymphoblastic leukemia with < 5% leukemic cells at laboratory testing. MRD was re-evaluated using dimensionality reduction (t-distributed stochastic neighbor embedding, Uniform Manifold Approximation and Projection, and Pairwise Controlled Manifold Approximation) implemented in a FlowJo plugin after pre-gating based on lineage and differentiation markers (CD19 or CD34/CD117). MRD was identified in 46 specimens, with leukemic cell percentages closely matching expert manual analysis, as well as in two specimens previously reported as MRD-negative after routine analysis, where abundant normal counterparts complicated conventional gating. Dimensionality reduction-based MFC-MRD analysis enables visualization of rare leukemic populations in single samples while maintaining concordance with expert gating, thereby facilitating broader adoption of highly sensitive flow cytometric MRD testing in clinical practice.