Yokosawa, D. | Search bioRxiv

bioRxiv2025

Citizen science gamers enable automated flow cytometry gating through machine learning

Manual flow cytometry gating requires up to one hour per sample with 32% inter-expert variability, creating critical bottlenecks in immunological research reproducibility. To address this, we developed flowMagic, a machine learning algorithm for automated gating that is trained on both expert-curated data (template data) and crowdsourced annotations from citizen science gaming. Through EVE Online, 839,199 players analyzed 52,178 bivariate plots from 37 studies, generating 31,703 quality-controlled training plots. Evaluated against 92,203 expert-validated files spanning 79 immune populations (i.e., a biologically defined cell subset within each bivariate plot), flowMagic achieved 90% accuracy for abundant populations and 65% for rare populations, outperforming existing methods. The algorithm reproduced biological patterns including neutrophil dynamics in COVID-19 patients and immune development in newborns. This gaming-based approach demonstrates that crowd-sourced pattern recognition generates robust training data for complex biomedical applications, offering transformative potential for standardizing flow cytometry analysis and accelerating immunological discovery.

bioinformatics↗

bioRxiv2025

Reproducible detection of antigen-specific T cells and Tregs via standardized and automated activation-induced marker assay workflows

Activation-induced marker (AIM) assays are a promising tool to track antigen-specific T cells, but methodological heterogeneity between research groups hinders their clinical utility. To evaluate AIM assay reproducibility, we conducted a multi-site study of SARS-CoV-2 and cytomegalovirus AIMs. We found inherent variability in AIM assays and optimized approaches to enhance reproducibility, including a standardized workflow to minimize technical variability and a generalizable Box-Cox transformation-based statistical method to optimize calculation of AIM stimulation responses. We further standardized AIM data analysis through development of automated flow cytometric gating software, which had superior reproducibility compared to manual analysis. We also characterized antigen-specific Tregs, finding that gating on a combination of CD134, CD137, FOXP3 and HELIOS optimally detects antigen-specific cells. The combined methodology results in a high degree of reproducibility within and between research groups and provides a comprehensive foundation from which standardized AIM assays can be implemented across diverse scientific and clinical settings. MOTIVATIONReliable detection of antigen-specific T cells is critical to understand immune responses to infection and vaccination, and has translational potential to monitor T cell responses across diverse clinical settings. Activation-induced marker (AIM) assays offer a variety of advantages over methods such as ELISPOT and tetramers, but are limited by methodological heterogeneity between research groups and a lack of standardized protocols. As such, the degree to which AIM assay results are reproducible is unknown. Key variables such as cell source, media, stimulation time, marker selection for CD4+ T cells, CD8+ T cells and regulatory T cells; as well as data analysis parameters such as flow cytometric gating strategies and mathematical correction for background AIM+ frequencies in unstimulated control samples, have not been rigorously studied. To address this, we comprehensively characterized variability in AIM assays, including within and between operators and across multiple research centres, and sought to optimize a standard AIM workflow to enhance reproducibility at both the experimental and analytical levels.

immunology↗