A formal Bayesian decisional framework for automated cell subset assignment in multiparametric flow cytometry: The...

Multiparametric flow cytometry (MFC) data analyses still largely rely on expert-dependent subjective manual gating strategies. While unsupervised clustering methods have improved data exploitation, no formal probabilistic decisional framework has yet been proposed for the critical subsequent step of cell subset assignment and quantitation. LBC-Flow (patent pending) is a structured naïve Gaussian Bayesian decisional framework built upon FlowSOM clustering. Unlike black-box machine learning approaches, every classification decision is fully interpretable, explicit and mathematically justified by its posterior probability value. A reference model including all cell distribution parameters must be constructed from normal body fluid, enabling the computation of posterior probabilities for each cell. Cells failing to meet predefined acceptance thresholds are flagged as a new specific class of Non-Classifiable Events (NCE). A leukocyte differential panel was used as proof-of-concept on 36 blood samples. Five analytical strategies were compared: manual gating, FlowSOM-only quantification, and three Bayesian classification approaches-(i.e. two discrete based on nodes (BDN) or cells (BDC) and one gaussian based on cells (BCGC)-. Using intraclass correlation coefficient ICC(3,1) and Z score for comparisons, BCGC demonstrated better performance than the other methods particularly for rare subsets. Complete results were delivered for 26 identified leukocyte subpopulations and NCE, in less than 30 s per sample. LBC-Flow is an original formal Bayesian decisional framework, explicitly addressing the gap between unsupervised clustering output and reproducible cell subset assignment and quantitation. Panel-agnostic by design, this approach provides a methodological foundation to be tested with other flow cytometry datasets in clinical or research contexts.