Fig. 4: Cell clustering and trajectory evaluations.
From: scGNN is a novel graph neural network framework for single-cell RNA-Seq analyses
a Comparison of ARI and Silhouette scores among scGNN and nine tools using Klein and Zeisel data sets. b Comparison of UMAP visualizations on the same two data sets, indicating that when scGNN embeddings are utilized, cells are more closely grouped within the same cluster but when other tools are used, cells are more separated between clusters. Cells were clustered via the Louvain method and visualized using UMAP. c Pseudotime analysis using the raw expression matrix and scGNN imputed matrix of the Klein data set via Monocle. d Justification of using the graph autoencoder, the cluster autoencoder, and the top 2000 variable genes on the Klein data set in the scGNN framework, in terms of ARI. scGNN CA- shows the results of the graph autoencoder’s ablation, CA- shows the results of the cluster autoencoder’s ablation, and AG shows the results after using all genes in the framework.
