Fig. 1: TACCO, a flexible framework for the annotation and analysis of cells and cell-like objects.

a, TACCO generates annotations for new datasets of mixtures (top left) using an annotated single-cell reference (top right) and provides methods for downstream analysis of the resulting compositional annotations (bottom left). b, Compositional annotations. Illustrative embeddings of cells and cell-like objects annotated (left) for mixtures (pie charts) of idealized pure contributions (triangles); (middle) as ambiguous annotations (triangles with colored borders) for technical artifacts like high ambient contributions or dropout levels and (right) continuous annotations (circles) along biological continua. c, Annotation process. Far left: a labeled reference dataset (for example, scRNA-seq data and colored triangles) and a new dataset (for example, Slide-seq beads and circles) are first presented in a common high-dimensional space (for example, expression space) optionally using platform normalization to make the datasets comparable. Near left: TACCO represents the reference categories by one or multiple representative profiles (large colored triangles). Near right: TACCO uses semi-unbalanced entropic optimal transport to transfer annotations from the reference categories to the new dataset (arrows), generating compositional annotations for the new datapoints (colored pie charts). To improve the capture of subdominant contributions, this process is iterated. Far right: TACCO provides compositional annotations for the new dataset. d, TACCO analysis tools for compositional annotations, especially for spatial data. From left: spatial relationship analysis on long (tissue) and short (cellular neighborships) length scales; inferring spatial regions by both spatial and annotation information; enrichment of compositional annotations and splitting compositionally annotated count data into pure contributions for downstream analysis with single-cell analysis tools.