Extended Data Fig. 8: DAAs are largely derived from Gfap-low astrocytes.

a-b, Direction of transition of AD astrocytes from their global optimal nearest neighbor (origin) cell (predicted by the Hungarian algorithm³⁴, Methods) to their given position in the diffusion map. For each cluster, force field (black arrows) marking the directionality of transitions along the diffusion map²⁸ (as in Fig. 1d), cells colored by cluster ID. Showing transitions from the predicted cell of origin among all WT cells or AD cells outside the cluster (a), or among all WT cells only (b) (transitions among AD cells only are shown in Fig. 3). c, Proportion of cells of origin per cluster (dot size and color) for each AD cluster (rows), from all WT cells (left) or from AD cells from all other clusters (right). d, Scheme of predicted transitions, shown as a graph with arrows between pairs of clusters with high proportion of origin cells (>15%) when mapping AD to all WT cells. Color and width of edge reflects the proportion. Diffusion map as in (a) in the background. e, Genes correlated with predicted transitions from WT cells to DAA in AD. The expression level across clusters (dot color) and the percent of cells expressing (dot size) significant (Pearson Correlation coefficient, FDR q-value<0.001, n=28 mice, 25,076 cells, Supplementary Table 1) genes that correlated with the transition to cluster 3 (intermediate) or cluster 4 (DAAs) from WT cells. Bottom: Assignment of each gene to a transition between pairs of clusters. f, Direction of transition on the diffusion space of AD astrocytes in each cluster from their weighted K-nearest neighbor position among all other WT cells. Force field showing the directionality of transitions between the expected position in the diffusion map (weighted average position of K-NN among all WT cells, k=10) to the true position along the diffusion map (as in Fig. 1d) for each cell. Colored by cluster IDs. Cluster numbers on top of each graph.