Fig. 4

CDI’s performance. CDI was applied to four simulated datasets (SD1-SD4) and three experimental datasets (T-CELL, CORTEX, RETINA); all have high-quality benchmark label sets. CDI-AIC was applied to the panels with (AIC) in the subtitles; CDI-BIC was applied to the rest of the panels. A CDIs of the candidate label sets. The x-axis labels the cluster numbers, and the y-axis labels the CDI scores. For the datasets with hierarchical structures (SD3, CORTEX, and RETINA), both CDI-BIC and CDI-AIC were applied. The line color represents the clustering method. The red triangle marks the CDI-selected label set; the purple star marks the benchmark main type label set; the brown star marks the benchmark subtype label set. B The heatmaps show the cell proportions from the benchmark label sets for each cluster in the selected label sets. The x-axis labels the CDI-selected label set clusters, and the y-axis labels the benchmark cell types. The color of each rectangle represents the benchmark cell-type proportions in each selected label set cluster. Each column adds up to 1. PR: photoreceptor