Fig. 3

The PDX somatic mutation calling workflow improved the accuracy of predicting somatic mutations in engrafted tumors. a Benchmarking of the CTP variant calling workflow using 45 simulated sequencing datasets from different samples, sequence coverages, and mouse DNA content (Additional file 1: Table S2) using precision, recall and F1 score (see Methods) based on the input variants for each sample. Complete: variant calling pipeline with all steps included; NoXenome: variant calling pipeline with Xenome omitted; all: all variants called by the pipeline; pass: variants annotated as “PASS” in the pipeline which pass the hard filters, minimum read depth and minimum alternate allele frequency of the variant. b Distribution of mutational load per sample of non-silent coding somatic mutations for genes included on the CTP panel based on exome sequencing of TCGA samples and CTP-panel sequencing of PDX models. TCGA somatic: TCGA somatic mutations reported in maf files; PDX: all variants annotated as “PASS” (pass the hard filters, minimum read depth and minimum alternate allele frequency of the variant); PDX filter germline: all variants annotated as “PASS” and filtered from putative germline variants; PDX filter germline & FP: all variants annotated as “PASS” and filtered from putative germline variants and recurrent false positives. (LUAD: lung adenocarcinoma, LUSC: lung squamous cell carcinoma, Colorectal: colon and rectal cancer, TNBC: triple-negative breast cancer, BLCA invasive: muscle invasive bladder cancer). c Mutations identified in patient lung tumors that were retained in engrafted PDX tumors. Some of these variants were initially filtered out of the variant call analysis but subsequently reinstated using the variant rescue protocol (Additional file 1: Figure S2)