Fig. 4: Analysis of Stereo-seq mouse embryo datasets with 8 development stages.
From: Prioritizing perturbation-responsive gene patterns using interpretable deep learning
A Dataset: Stereo-seq mouse embryo dataset across eight development stages ([E9.5, E10.5, E11.5, E12.5, E13.5, E14.5, E15.5, E16.5]). B Visualization of River-identified top-5 genes using count expression values. C Visualization of different gene set inputs (2000 HVG genes, River-selected top-20 genes, and two versions of River-selected bottom genes (DE and SVG) (see Methods)). Notably, the SVG bottom genes’ embedding is collapsed in the embedding space due to the lack of information, providing a negative control for River-selected genes. D Unsupervised clustering results comparison between different input gene sets (top-5 genes, 2000 HVG genes, and River-selected bottom genes (two versions)) using NMI, ARI, and cLISI metrics. E Pairwise silhouette score across eight development stages for different input gene sets. The pairwise silhouette score reflects the distance between two clusters in the UMAP space in Fig. 4C. F Principal Component Analysis for River-selected top-5 genes colored by development stages. The right panel shows the principal components’ distribution change tendency along with developmental changes. Box plots display the principal component values of individual cells across developmental stages (n = 5000 point per development stage; each point represents one cell’s principal component score). Box plots indicate the median (center lines), interquartile range (hinges), and 1.5× interquartile range (whiskers). G Visualization of River-identified top genes using binary expression values (unique to the gene set of River-count). H Significant enriched (FDR adjusted p-value < 0.05) gene sets uniquely identified by River-binary in GO biological process reference.
