Assessing the clinical applicability of dimensionality reduction algorithms in flow cytometry for hematologic malignancies

Objectives: Despite its utility, interpreting multiparameter flow cytometry (MFC) data for hematologic malignancy remains time-intensive and complex. This study evaluated the applicability of two dimensionality reduction (DR) algorithms, t-distributed stochastic neighbor embedding (t-SNE) and uniform manifold approximation and projection (UMAP), to MFC data of hematologic malignancy. Methods: A total of 237 samples were re-analyzed by t-SNE- and UMAP-based gating: 80 with acute leukemia orientation tube panel, 42 with B-cell lymphoma (BCL) panel, 45 with multiple myeloma (MM) panel, 40 and 30 with measurable residual disease (MRD) panels for B-cell acute lymphoblastic leukemia (B-MRD) and MM (MM-MRD), respectively. Each result was compared to the manual gating, and sensitivity and precision were assessed using BCL and B-MRD panels. Results: Compared to manual gating, DR-based gating demonstrated agreements over 95.0% for all MFC panels, and quantitative correlations (ρ) exceeded 0.94. Both t-SNE- and UMAP-based gating showed a sensitivity and negative predictive value of 100%. Also, in one sample each from the BCL and MM-MRD panels, DR-based gating identified populations that were missed by manual gating. Sensitivity evaluation showed that both t-SNE- and UMAP-based gating successfully identified MRD populations down to the lowest MRD level of 10-5.30 when applying primary-gating strategy for CD19-positive population. Precision evaluation showed coefficient of variation below 10% across all levels. Conclusions: This study shows that DR-based gating streamlines data interpretation and minimizes overlooked populations, demonstrating significant potential as a valuable tool in MFC analysis for hematologic malignancies.