Clustering-Based Rare-Event Detection Beyond Singlets in Flow Cytometry
Kwak, Yejin 1; Würth, Roberto; Yamachui Sitcheu, Angelo Jovin
1; Hübschmann, Daniel; Mikut, Ralf
1
1 Institut für Automation und angewandte Informatik (IAI), Karlsruher Institut für Technologie (KIT)
1; Hübschmann, Daniel; Mikut, Ralf
11 Institut für Automation und angewandte Informatik (IAI), Karlsruher Institut für Technologie (KIT)
| Zugehörige Institution(en) am KIT | Institut für Automation und angewandte Informatik (IAI) |
| Publikationstyp | Proceedingsbeitrag |
| Publikationsjahr | 2026 |
| Sprache | Englisch |
| Identifikator | KITopen-ID: 1000198482 |
| Erschienen in | Proceedings. 36. Workshop Computational Intelligence : Berlin, 19. – 20. November 2026. Ed.: H. Schulte |
| Veranstaltung | 36. Workshop Computational Intelligence (2026), Berlin, Deutschland, 19.11.2026 – 20.11.2026 |
| Verlag | KIT Scientific Publishing |
| Seiten | 283-305 |
| Bemerkung zur Veröffentlichung | Physical cell interactions are biologically relevant, but they are routinely discarded as doublets or multiplets in standard flow cytometry workflows. A recent clustering-based cytometry framework has instead retained multiplets while using non-uniform sampling, or sketching, to preserve rare cell types during large-scale analysis. However, it remains unclear how well such sampling strategies support downstream recovery of extremely rare events and how reliably singlets can be distinguished from multiplets. To address this, we compare leverage-score and geometric sketching for recovering circulating tumor cells (CTCs) among singlet events, and evaluate whether adding side scatter (SSC) ratio to forward scatter (FSC) ratio improves singlet/multiplet discrimination across all events. The analysis comprises 4 blood samples from 3 patients, ranging from 116,507 to 1,361,065 events each. Geometric sketching yielded higher CTC recovery on average, particularly in recall, though performance remained sensitive to sample identity and clustering resolution. FSC ratio alone provided strong singlet/multiplet separation, while adding SSC ratio yielded no consistent improvement. These findings highlight the importance of both sampling strategy and feature selection for extending clustering-based cytometry to extremely rare events while retaining information beyond singlets. |
| Relationen in KITopen |