Fig. 1

Overview of the scDrugPrio workflow. Samples were derived from inflamed tissue in patients (red) and healthy tissues in controls (blue) or, for individual predictions, sampling included inflamed and uninflamed tissue taken from the same patient. Single-cell RNA-sequencing (scRNA-seq) data were preprocessed by undergoing quality control, denoising, clustering, cell typing and differentially expressed gene (DEG) calculation. DEGs for each cell type were calculated between healthy and sick samples. Using DEGs alongside information on drugs, scDrugPrio selects drug candidates (for each cell type; CT) whose gene targets are (1) in network proximity to DEGs (network proximity based selection) and (2) who counteract disease-associated expression changes (pharmacological action filtering). These cell type-specific drug candidates are next ranked using intracellular and intercellular centrality. (3) Intracellular centrality is computed based on the centrality of drug targets in the largest connected component (LCC) formed by DEGs and functions as a proxy for drug target importance. (4) Intercellular centrality measures centrality in disease-associated cellular crosstalk networks called multicellular disease models (MCDMs). (5) To derive a final ranking that aggregated cell type-specific drug selection and ranking into one list, drug candidates were ranked using a composite score of intra- and intercellular centralities (Additional file 1: Fig. S1). Drugs with identical targets and mechanism of action were given identical rankings