Tribus: Semi-automated discovery of cell identities and phenotypes from multiplexed imaging and proteomic data

• DOI: 10.1101/2024.03.13.584767
• Publication Date: 2024-03-14T10:45:22Z
• Authors: Ziqi Kang; Angela Szabo; Teodora Farago; Fernando Perez-Vi...; Ada Junquera; Saundarya Shah; Inga-Maria Launonen; Ella Anttila; Julia Casado; Kevin Elias; Anni Virtanen; Ulla-Maija Haltia; Anniina Färkkilä
• Keywords: 11832 Microbiology and virology; Single; Medical biotechnology; Mass cytometry

Abstract Motivation Multiplexed imaging and single-cell analysis are increasingly applied to investigate the tissue spatial ecosystems in cancer other complex diseases. Accurate phenotyping based on marker combinations is a critical but challenging task due (i) low reproducibility across experiments with manual thresholding, and, (ii) labor-intensive ground-truth expert annotation required for learning-based methods. Results We developed Tribus, an interactive knowledge-based classifier multiplexed images proteomic datasets that avoids hard-set thresholds labeling. demonstrated Tribus recovers fine-grained cell types, matching gold standard annotations by human experts. Additionally, can target ambiguous populations discover phenotypically distinct subtypes. Through benchmarking against three similar methods four public ground truth labels, we show outperforms accuracy computational efficiency, reducing runtime order of magnitude. Finally, demonstrate performance rapid precise two large in-house whole-slide datasets. Availability available at https://github.com/farkkilab/tribus as open-source Python package.