Extended Data Fig. 2: Comparison of WT MEEB to in-vivo gastrulation atlases.

a, Shown are color coded expression level of key transcription factors over metacells (columns) that were derived from the Pijuan-Sala et al gastrulation atlas data set. Lower colored labels are based on the annotation in the original analysis. b, Shown are metacells (colored ovals) projection in 2D using the metacell 2D graph projection, illustrating the reference in-vivo atlas we used for initial annotation of MEEB models. c, Each cell in the MEEB data set was tested for correlation with each metacell profile in the atlas, deriving a putative cell type annotation from the best matching atlas state. Shown are the distribution of annotations in each MEEB metacell (bottom), and the entropy of this annotation, which we use to threshold the decision on MEEB metacell cell type annotation. d, For each MEEB metacell we compared the MEEB UMI distribution to a distribution derived by pooling atlas RNA according to the best matching atlas metacell for each MEEB single cell (Methods). Shown is the correlation between MEEB and pooled atlas RNAs for each metacell. e, Scatter plots compare the absolute expression (log2 Umi frequency) in MEEB metacells and the expression expected given projection on the MARS-seq temporal mouse gastrulation atlas. Good quantitative matching is observed for multiple genes, suggesting the combinatorial transcriptional state in MEEB cells is highly similar to the in-vivo state. Genes showing imperfect scaling of expression in a subset of the metacells represent in-vivo/in-vitro differences that must be considered carefully. Note increased Dnmt3b expression in gut and increased Dnmt3a expression in mesenchyme, accentuating the trends observed in vivo (Fig. 1a).