Extended Data Fig. 9: Additional performance evaluation metrics in the Centralized dataset analysis and across state-of-the-art data integration approaches.
From: Interpretable inflammation landscape of circulating immune cells
(a) Pointplot showing the Balance Accuracy Score (top), and Matthew Correlation Coefficient (bottom) computed, considering Majority Vote, 100 random disease assignments, and cell type prediction, on the samples from left out pools in the Centralized Dataset. (b) Heatmap reporting Recall and Precision obtained on the samples from left out pools by each cell type for each disease included in the centralized dataset. (c-d) Performance evaluation from Scenario 2 (c) and Scenario 3 (d), respectively, showing (left) the distribution of Weighted Recall and Weighted Precision for all the configurations of each data integration approach, and (right) the mean and standard-deviation of each data integration method, including 100 random label assignments. Arrows highlight the scANVI configuration applied in Scenario 1.
