Cell-cell interaction analysis

We applied CellphoneDB’s statistical analysis method (2.1.2) and receptor-ligand database (2.0.0) to calculate statistically enriched cell-cell interactions (https://github.com/Teichlab/cellphonedb). We used the log-normalized RNA assay of our BALF dataset containing all samples and selected cells either from patients with COVID-19 or patients with bacterial pneumonia to gain the count matrix. Because we have different levels of subclustering, we annotated each cell according to its cluster in its deepest level and used it as metadata input (clustering level: all BALF samples > T cells > CD4+ T cells; all BALF samples > myeloid cells). We ran CellphoneDB with the default parameters. In total, CellphoneDB returned 13,034 significant (P < 0.05) interactions. The rank of a ligand-receptor pair was calculated by CellphoneDB dividing its total number of significant P values by the number of cluster-cluster comparisons. For downstream analysis, we excluded integrin ligand-receptor pairs and interactions being annotated as not secreted. Moreover, we excluded CCL20-CXCR3 interactions (Id_cp_interaction “CPI-SS0F8C664D9”) because of a lack of evidence in the literature.