(168) A Method to Detect Differences in Heterogeneous Cell Populations Without Assuming a Specific Data Distribution leveraging SOM-based clustering

Introduction/Rationale: Flow cytometry (FCM) is often used to identify specific cell populations that exhibit changes in abundance or count under different conditions. Recent advances in high-dimensional FCM have increased the number of simultaneously measurable parameters, making manual gating impractical and leading to the use of clustering-based analysis to detect differential abundance in cell populations across groups. There are several challenges with its current usage and methodology. First, even though the underlying distribution model of cluster frequencies is typically unknown, parametric statistical tests such as t-test are widely used. Second, existing methods do not adequately account for within-group heterogeneity. These challenges result in user-dependent trial-and-error in statistical analysis.

Methods: To address these issues, we developed a new differential abundance method for comparing two groups using a non-parametric test, simultaneously quantifying within-group heterogeneity. The concatenated sample data from the two groups are analyzed using SOM-based clustering to construct a sample × cluster matrix, which is then decomposed by Nonnegative Matrix Factorization (NMF). Differentially abundant clusters are identified with basis matrix via a permutation test, and within-group heterogeneity are quantified with coefficient matrix.

Results: We applied our method to 45-color human PBMC datasets acquired using a Sony ID7000™ Spectral Cell Analyzer (n=8). Compared with manual analysis, our approach successfully detected differentially abundant clusters, which were also noted by manual analysis. In addition, by quantitatively evaluating within-group heterogeneity at the sample level, our method identified samples with intermediate or unanticipated patterns of cell abundance, which were discussed in manual analysis.

Conclusion: This approach has the potential to streamline the multi-sample analysis workflows and enhance the interpretability of differential abundance analyses in FCM research.