Fig. 4: GEMLI correctly assigns cells to individual metastases, crypts, and mitochondrial variant-defined lineages in vivo.
From: Gene-expression memory-based prediction of cell lineages from scRNA-seq datasets
a Scheme of GEMLI lung metastasis predictions. b Precision (line) and sensitivity (dotted line) as a function of confidence level for GEMLI predictions in lung metastases of two mice tested against the CRISPR barcode lineage information. Line: mean; Shade: S.D. c Pie chart of the lung metastases data of two mice (M1, M2) as in b colored according to the presence of CRISPR barcode- (left) and GEMLI- lineage annotation at a confidence level of 50 (middle) and 80 (right). The number of recovered lineages is indicated. Conf=confidence level. d Schematic of GEMLI crypt predictions. e Percentage of cells predicted as individual lineages for each crypt in the crypt dataset (n = 1) at confidence level 70. f Predictions of crypts at confidence level 70. g Precision-sensitivity curve for predictions in the crypt dataset. h Precision-sensitivity curve for predictions of cell pairs composed of one cell type (symmetric) or two cell types (asymmetric) in the crypt dataset. i The percentage of cells (size) belonging to the indicated cell type (color) for individual crypts (top) and for predicted lineages at confidence level 50 (bottom) aligned to their crypt of origin. j Precision-sensitivity curves of lineage predictions in bone marrow cells using mitochondrial variants as ground truth lineages (n = 3 datasets). Line: mean; Shade: S.D. k Precision-sensitivity curves of lineage predictions in K562 cells (n = 1 dataset) as in j. l Precision-sensitivity curves of lineage predictions in bone marrow cells using mitochondrial variants as ground truth lineages as in j with lineage size parameters up to the maximal ground truth lineage size. Line: mean; Shades: S.D. All crypt predictions are of 5-40 cells and were run with a lineage size parameter of 2–40. Source data are provided as a Source Data file.
