Fig. 4: Ablation model experiment results and the dimensionality reduction plots of cells and embeddings from the GSE117872 dataset. | Communications Biology

Fig. 4: Ablation model experiment results and the dimensionality reduction plots of cells and embeddings from the GSE117872 dataset.

From: scXDR: drug response prediction across single-cell datasets via heterogeneous network transfer learning

Fig. 4: Ablation model experiment results and the dimensionality reduction plots of cells and embeddings from the GSE117872 dataset.

a The percentage decrease in metrics for each ablation model on the combined scenario, calculated by the median of each metric. b First row: UMAP dimensionality reduction of cells’ true response labels, Louvain clusters, cell types, and predicted response values. Second row: PCA dimensionality reduction of cells’ original features, embeddings after the scXDR message passing phase, embeddings after alignment in the scXDR transfer learning phase, and the scXDR score scatter plot. Third row: PCA dimensionality reduction plots of source and target domains for scDEAL (bulk) and SCAD (bulk), along with score scatter plots. Fourth row: PCA dimensionality reduction plots of source and target domains for scDEAL and SCAD, along with score scatter plots. Red text refers to the Euclidean distance between the mean points of the source and target domains. In the first row, the x and y axes of all plots represent UMAP1 and UMAP2. In rows 2 to 4, the x and y axes of the PCA plots represent Principal Components 1 and 2, and in the score scatter plots, the x-axis corresponds to the cell index and the y-axis corresponds to the score.

Back to article page