Unsupervised Analysis of Flow Cytometry Data in a Clinical Setting Captures Cell Diversity and Allows Population Discovery

Abstract

Data obtained with cytometry are increasingly complex and their interrogation impacts the type and quality of knowledge gained. Conventional supervised analyses are limited to pre-defined cell populations and do not exploit the full potential of data. Here, in the context of a clinical trial of cancer patients treated with radiotherapy, we performed longitudinal flow cytometry analyses to identify multiple distinct cell populations in circulating whole blood. We cross-compared the results from state-of-the-art recommended supervised analyses with results from MegaClust, a high-performance data-driven clustering algorithm allowing fast and robust identification of cell-type populations. Ten distinct cell populations were accurately identified by supervised analyses, including main T, B, dendritic cell (DC), natural killer (NK) and monocytes subsets. While all ten subsets were also identified with MegaClust, additional cell populations were revealed (e.g. CD4+HLA-DR+ and NKT-like subsets), and DC profiling was enriched by the assignment of additional subset-specific markers. Comparison between transcriptomic profiles of purified DC populations and publicly available datasets confirmed the accuracy of the unsupervised clustering algorithm and demonstrated its potential to identify rare and scarcely described cell subsets. Our observations show that data-driven analyses of cytometry data significantly enrich the amount and quality of knowledge gained, representing an important step in refining the characterization of immune responses.

Time series clustering of T cell subsets dissects heterogeneity in immune reconstitution and clinical outcomes among MUD-HCT patients receiving ATG or PTCy

“…However, this canonical analysis of immune reconstitution focuses on the examination of one cell subset at a time not reflecting the interplay between distinct cellular subsets. Here, the use of median values may be efficient in providing an overview of cellular reconstitution ( 52 ) for specific patient subsets but are not very conclusive about the individual patient. This limitation may be overcome using the approach of time series clustering of multi-dimensional flow cytometry data, which to our knowledge has not been published before.…”

Investigating Vα7.2+/CD161− T Cell and MAIT Cell Profiles Using Flow Cytometry in Healthy Subjects and Subjects with Atopic Dermatitis

“…However, it has been verified that machine learning algorithms avoid manual biased gating and potentially detect novel cell types and cellular relationships. These populations might be missed in traditional gating due to the complexity of cellular heterogeneity and the limitation of exploring all the dimensions of datasets at the same time [ 38 , 39 , 40 , 41 ]. In our case, the additional phenotypes that we identified using the multidimensional analysis were CD8 + subsets of Vα7.2 + /CD161 − T cells; CD69 + , CD4 + , CD8 + , and DN MAIT cells; and CD161 + and CD4 + /CD161 + T cells.…”

Analysis of Antigen‐Specific T and B Cells for Monitoring Immune Protection Against SARS‐CoV‐2

“…A large cytometric sample can result in inefficient coverage in the detection of a number of spurious small populations (often outliers of larger, noisy populations) (Qi et al, 2020). Moreover, tuning the parameters of the analysis could be very effective for rare populations (Baumgaertner et al, 2021).…”