Fig. 6

scDrugPrio for individual drug prediction. a, b Cell type proportions differed greatly between two CD patients, as shown in the horizontally stacked bar plots representing paired biopsies from inflamed (infl) and uninflamed (uninfl) lesions that were taken from each patient. c, d Patients also show differences in the composition and interconnectivity (representing ligand interactions) of the MCDMs. Patient 1 had a cell type for which no ligand-target interactions could be found with any other cell types in the MCDM. e, f Precision among ranked candidates for approved CD drugs, drugs for which clinical trials had been registered for CD and literature evidence (only for top 100 ranking drugs). The precision for the ranked drug candidates for patient 1 was low for approved CD drugs, while literature evidence supported the top-ranking drugs, of which many are anti-inflammatory. In contrast, patient 10 had several approved CD drugs among the top 100 candidates, and a curated literature search confirmed the validity of many more candidates. g Precision for prediction based on pooled patient data was poor. h Venn diagram presenting the overlap of considered drug candidates for patients 1 and 10. i Interindividual differences between patients 1 and 10 were reflected in the prediction outcome, as no correlation existed between the drug rank of drugs that were candidates in both patients