Fig. 5: NMF captures transcriptional programs relevant to neuronal activity.

a, Select NMF patterns projected onto a mouse snRNA-seq dataset of hippocampal neurons activated by ECS or under control conditions (Sham; y axis, by cell type). The dot color indicates scaled average nuclei NMF weights. Dot size indicates the proportion of y axis group with non-zero pattern weight for the given x axis value. DE analysis was performed on mouse GC nuclei, testing for differences in the expression by activity condition. For all volcano plots (b,c,f,g), the y axis presents the −log₁₀ FDR-adjusted P value and the x axis presents log₂(FC), where negative values indicate greater expression in sham-activated GCs and positive values indicate greater expression in ECS GCs. b,c,f,g, Points are colored for the gene-level NMF weight for nmf91 (b), nmf20 (c), nmf10 (f) and nmf14 (g). d,e, Spot plots isolating the DG GCL spatial domain (green outlined spots) demonstrate the differing spatial organization of nmf10 (d) and nmf14 (e) weights in an example capture area from donor Br3942. Spot fill indicates spot-level NMF pattern weight. h, Left, UMAP plot of all nuclei present in our human snRNA-seq, highlighting the GC clusters. Right, zoomed UMAP plot of only GC nuclei from our human snRNA-seq dataset with color indicating cluster identity. i, UMAP plot of human GC nuclei showing (left) nmf10 nuclei-level weights and (right) log₂-normalized counts of highly-weighted nmf10 gene CHST9. j, UMAP plot of human GC nuclei showing (left) nmf14 nuclei-level weights and (right) log₂-normalized counts of highly-weighted nmf14 gene SORCS3. PS/Sub, prosubiculum and subiculum neurons; L5/Po, layer 5 and polymorphic layer.