Automated mapping of phenotype space with single-cell data

Abstract

Accurate and rapid identification of cell populations is key to discovering novelty in multidimensional single cell experiments. We present a population finding algorithm X-shift that can process large datasets using fast KNN estimation of cell event density and automatically arranges populations by a marker-based classification system. X-shift analysis of mouse bone marrow data resolved the majority of known and several previously undescribed cell populations. Interestingly, previously known cell populations, as well as intermediate cell populations in early hematopoietic development, were described via novel marker combinations that were defined via routes to their locations in expressed marker space. X-shift provides a rapid, reliable approach to managed cell subset analysis that maximizes automation that not only best mimics human intuition, but as we show provides access to novel insights that “prior knowledge” might prevent the researcher from visualizing.

High-resolution myogenic lineage mapping by single-cell mass cytometry

“…Single-cell suspensions of hindlimb muscles, Gastrocnemius (GA) and Tibialis Anterior (TA) were prepared from 8-week old mice as described previously 10 and processed for CyTOF analysis (Fig.1b). To distinguish stem and progenitor cell populations, we analyzed the CyTOF dataset with a recently developed K-nearest neighbor density-based clustering algorithm called X-shift 13 , which performed unsupervised clustering analysis of cells within the myogenic compartment, defined as Lineage − /α 7 integrin + cells and subsequently refined to Lineage − /α 7 integrin + /CD9 + cells. With this new multivariate algorithm six clusters were generated, based on the expression of known and previously unrecognized surface markers, and key myogenic transcription factors.…”

“…With this new multivariate algorithm six clusters were generated, based on the expression of known and previously unrecognized surface markers, and key myogenic transcription factors. In order to visualize the spatial relationships between the cell types within these X-shift clusters, 2000 randomly sampled cells from each cluster were subjected to a force-directed layout 13–15 . The resultant map revealed that cells fell densely into three prominent clusters and were linked by sparsely populated paths (Fig.1c).…”

“…These clusters were visualized using single-cell force-directed layout. Up to 2000 cells were randomly selected from each X-shift cluster, each cell was connected to 30 nearest neighbors in the phenotypic space and the graph layout was generated using the ForceAltas2 algorithm 13–15 (representative experiment, n= 3 mice; 4 independent experiments). (d) Expression level of the myogenic transcription factors Pax7, Myf5, MyoD and Myogenin was visualized in the X-shift clusters shown in (c).…”

Single-cell developmental classification of B cell precursor acute lymphoblastic leukemia at diagnosis reveals predictors of relapse

“…2c). We found that Mahalanobis distance was superior to other distance metrics (cosine 14,16 , Euclidian 17 , and Manhattan 18 ) in assigning cells to the correct developmental population (Supplementary Fig. 2d).…”

Automated cell type discovery and classification through knowledge transfer

“…Three experts independently gated the cellular populations in the PANORAMA dataset and only the consensus part of the gating was retained. All event measurements were transformed by ${\sinh }^{-1}\left(\right(x-1)/5)$before further processing (Samusik et al , 2016). …”