Extended Data Fig. 4: Identifying shared spatial patterns of gene expression using non-negative matrix factorization.

(A) Illustrations of the definition of ‘spatial bins’ used in gene expression analyses (purple outlines) and ‘cubelets’ used in the analyses of H3 type distributions (blue outlines) in the pilot brain. The definition of spatial bins aimed for equal cell numbers across bins within a slice, whereas the definition of cubelets aimed for equal width on the surface of the cortex. Dots indicate cells. (B)(C) Variance in gene expression explained by space compared to that explained by H2 types (B) or additional variance explained by H3 types (C). In (B) dashed lines indicate threshold for p = 0.05 (one-sided F-test with K-1 and 21,197-K degrees of freedom, where K = 8 H2 types or K = 540 spatial bins, Bonferroni corrected for multiple comparisons). (D) The expression patterns of the indicated genes (Ctgf, Nnat, and Tenm3) plotted on flatmaps of the cortex in all cells (left column) or in each H2 type (center). The variations of gene expression in each H2 type along the AP axis and the ML axis are shown on the right. Line colors indicate H2 types as shown in the center plots. In the upper left plot, the same map is color-coded and labeled by cortical areas. (E) Spatial patterns of all 10 NMF factors. (F) Cumulative variance explained by the indicated number of NMF factors. (G) Spearman correlation between NMF factors and cortical areas. (H) Histogram of the number of NMF factors that each gene is associated with.