Figure 3.

External validation of the disease specificity of pancreatic ductal adenocarcinoma (PDAC) faecal microbiome models. False positive rate (FPR) of metagenomic unconstrained model-1 and enrichment-constrained model-2 in 25 external test sets is shown as a bar plot (see online supplemental table S4 for a list of all studies included). Validation datasets were profiled and normalised in the same way as the initial dataset (see ‘Methods’). Each study was stratified according to health status and models were tested to predict in the given group at a 90% specificity cut-off. A low FPR on metagenomes from patients with other disorders and healthy individuals indicates that the model is specific to PDAC. The number of subjects in each group is displayed as colour coded circles below. BRCA, breast cancer; CRC, colorectal cancer; CD, Crohn’s disease; CP, chronic pancreatitis;, CTR, controls; LD, liver disease; NAFLD, non-alcoholic fatty liver disease; PC, pancreatic cancer; T1D, type 1 diabetes; T2D, type 2 diabetes; UC, ulcerative colitis; ES, Spanish; DE, German.