HAL-X: Scalable hierarchical clustering for rapid and tunable single-cell analysis
Fig 6
Analysis of Lupus patients’ blood profile using single-cell mass cytometry and HAL-x.
(A) The pipeline for profiling immune cells from Lupus patients and healthy donors, beginning with blood collection, continuing with mass cytometry, and culminating with HAL-x processing. (B) t-SNE visualizations of clusterings generated by HAL-x which correspond to various levels of population specificity (cv-score). (C) Heatmap of the average expression levels for the 19 leukocyte clusters defined by HAL-x for a cv-score of 0.7. The Cell type annotations were made hand. (D) A graph of AUC (i.e. Area under the Receiver Operating Curve (ROC)) for different cv-scores shows that the most important clusters with discriminatory power are not observed for high CV-scores. The best cluster identified by HAL-x (cluster #42 for cv-score of 0.7), separates individual neutrophils from Lupus and Healthy patients with an AUC of 0.82. (E) Markers on neutrophils (cluster #42) that can be used to distinguish blood samples from Lupus patients and from healthy donors. Note how subtle the difference is at the level of individual markers.