Figure 4. Monocyte gene modules and associated cytokines are predictive of COVID-19 mortality.

a) PCA of COVID-19 patients and their outcomes based on down-selected immune gene modules scores derived from scRNAseq of PBMC samples (day 1 ICU). b-c) Area under the curve (AUC) scores derived from leave-one-out cross-validation using random forest (b) or support vector machine (c) algorithms. The diagonal line in each AUC plot represents a model with no predictive power (i.e. a 50% probability of death or survival). Random forest and support vector machine models achieved AUC scores of 0.79 and 0.88, respectively. d) Correlation plot of Mono_cells_PDE6H gene module scores with levels of MIP-1β in plasma on day 1. e) Proportional hazards regression analysis of MIP-1β levels on day 1 and mortality. f) Correlation plot of Mono_cells_NOC2L gene module scores with levels of CXCL10 in plasma on day 1. g) Proportional hazards regression analysis of CXCL10 levels on day 1 and mortality. h-k) Correlation plots of indicated monocyte modules using day 1 and day 5 scores. Mortality outcomes of COVID-19 patients analyzed by such longitudinal sampling are indicated by colored dots. I) Heatmap displaying correlation analysis of Mono_cells_NOC2L gene module scores with a network of inflammatory cytokines in plasma on day 5 in COVID-19 samples.