Supplementary Information Retrospective evaluation of whole exome and genome mutation calls in 746 cancer samples Bailey et al. Supplementary Figure 1 Non-hypermutators (< 1000 total mutations) a BLCA BRCA CESC COAD DLBC b Fraction matched MC3 only PCAWG only 0.6 1.00 1.00 1.00 0.4 Fraction of mutations 0.2 0.0 0.75 0.75 0.75 Fraction of unique PCAWG variants GBM HNSC KICH KIRC KIRP 0.6 0.4 0.50 0.50 0.50 0.2 0.0 LGG LIHC LUAD LUSC OV 0.25 0.25 0.25 0.6 0.4 0.2 0.00 0.00 0.00 0.0 0 1 0 1 0 1 PRAD READ SARC SKCM STAD c Complete barcode match = 1, partial match = 0 0.6 0.4 0.2 PCAWG (n=8185) 33.7% 66.3% 0.0 0 1 0 1 0 1 THCA UCEC MATCH (n=83075) 37.5% 62.5% 0.6 0.4 MC3 (n=11929) 42.5% 57.5% 0.2 0.0 100% 80% 60% 40% 20% 0% 20% 40% 60% 80% 100% 0 1 0 1 Complete barcode match = 1, partial match = 0 Partial ID match Exact ID match Including hypermutators d BLCA BRCA CESC COAD DLBC e Fraction matched MC3 only PCAWG only 0.6 1.00 1.00 1.00 0.4 Fraction of mutations 0.2 0.0 0.75 0.75 0.75 Fraction of unique PCAWG variants GBM HNSC KICH KIRC KIRP 0.6 0.4 0.50 0.50 0.50 0.2 0.0 LGG LIHC LUAD LUSC OV 0.25 0.25 0.25 0.6 0.4 0.2 0.00 0.00 0.00 0.0 0 1 0 1 0 1 PRAD READ SARC SKCM STAD f Complete barcode match = 1, partial match = 0 0.6 0.4 0.2 PCAWG (n=25639) 18.9% 81.1% 0.0 0 1 0 1 0 1 THCA UCEC MATCH (n=155860) 26.5% 73.5% 0.6 0.4 0.2 MC3 (n=21626) 29.0% 71.0% 0.0 0 1 0 1 100% 80% 60% 40% 20% 0% 20% 40% 60% 80% 100% Complete barcode match = 1, partial match = 0 Partial ID match Exact ID match g h PCAWG concordance (625 samples this study) MC3 concordance (625 samaples this study) 1.1 1.0 1.0 GBM BLCA SKCM KIRP OV SKCM BLCA KIRC CESC LUSC LUSC 0.9 KIRP DLBC STAD 0.9 BRCA LIHC COAD HNSC LGG UCEC BRCA COAD HNSC UCEC STAD GBM CESC DLBC LIHC LUAD SARC KIRC SARC 0.8 KICH LGG LUAD 0.8 OV THCA 0.7 THCA KICH Mann−Whitney p−value=0.000659 0.7 Mann−Whitney p−value=0.00197 Pearson’s p−value=0.00873 Pearson’s p−value=0.0558 0.6 Pearson’s R2=0.31 Pearson’s R2=0.179 PRAD PRAD 0.6 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 Non−Syn Muts per MB (Calculated from 8852 TCGA samples) Non−Syn Muts per MB (Calculated from 8852 TCGA samples) Supplementary Figure 1 Supplementary Figure 1: Effects of exact barcode matching on mutations concordance. a) Boxplots are calculated for fraction of mutations that are unique to the PCAWG data set. The fraction represents the number of mutations per sample, including MATCHED and MC3 unique variant calls. Here, we com- pare fractions of mutations by cancer types. b) Pan-Cancer fractions for MATCHED, MC3 unique and PCAWG unique variants are shown. Fractions are calculated as above. c) A Likert plot is used to assess the fraction of variants from the same TCGA barcode (same plate location, green color) to other vials, portions, or analytes of the same tumor (orange color). The full bar represents 100% of the fraction and is divided at the midline. The number of variants for each category is shown on the y-axis. Here, only non-hypermutated samples are shown to avoid confounding interpretation. Hypermutators are defined as samples with a total mutation count > 1000; this result shows n=720. d-f) Same as above, but with hypermutated samples (n=744). Points indicated on panels a, b, d, e indicate samples that extend past 1.5 times the interquartile range. g) Correlation plot of MC3 concordance statistics and non-silent muta- tions per megabase. h) Like the previous panel, this panel illustrates the correlation between PCAWG concordance percentages and non-silent mutations per megabase of the larger TCGA cohort. Green shading indicates 95% CI (panels h and g). Supplementary Figure 2 60000 15000 60000 40000 40000 10000 30000 40000 20000 20000 5000 20000 10000 0 0 0 0 Bladder−TCC Bone−Leiomyo Breast−AdenoCa Breast−LobularCa 250000 2500000 40000 200000 2000000 75000 30000 150000 1500000 50000 20000 100000 1000000 25000 50000 500000 10000 0 0 0 0 Count CNS−GBM ColoRect−AdenoCA Head−SCC Kidney−RCC PCAWG 200000 MC3 125000 60000 60000 150000 100000 75000 40000 40000 100000 50000 20000 50000 20000 25000 0 0 0 0 Liver−HCC Lung−AdenoCA Lymph−BNHL Ovary−AdenoCA 5e+05 3e+05 20000 750000 4e+05 15000 2e+05 500000 3e+05 10000 2e+05 1e+05 5000 250000 1e+05 0 0 0e+00 Prost−AdenoCA Skin−Melanoma Stomach−AdenoCA Uterus−AdenoCA Supplementary Figure 2: Effects of supplementing PCAWG samples with TCGA samples by cancer type. Mutations counts of cancers types split by TCGA and the rest of PCAWG samples. Box-plots show the number of mutations for all PCAWG samples and split by the portion that are also TCGA. Points indicate samples that extend past 1.5 time the interquartile range. Horizontal bars shows me- dian mutation counts. Here we show that mutations counts do not differ greatly between cohorts and that mainly the excess of mutations after sample selection is due to a few outliers and not cancer types selection bias. Supplementary Figure 3 a b 156000 68.4% 31.6% StrandBias (n=320) Retained wga & 41.4% 58.6% 152000 common_in_exac MATCH variant counts removed oxog artifacts (n=1,100) 21.5% 78.5% nonpreferredpair 148000 Mutations (n=1,114) count 16.5% 5000 83.5% wga Removed wga 4000 (n=15,831) 144000 & oxog artifacts 3000 14.9% 85.1% 2000 native_wga_mix 1000 (n=308) 18000 19000 20000 21000 93.3% 6.7% gapfiller Unique MC3 variant count (n=223) MC3 only 100% 80% 60% 40% 20% 0% 20% 40% 60% 80% 100% PCAWG & MC3 Supplementary Figure 3: Assessing MC3 filter flags. a) Displays results from an exhaustive search for an optimal set of MC3 filters. We performed an exhaustive search of MC3 filter combinations to identify an optimal set of MC3 filters to apply to our overlap criteria. This effort displayed using a resulted to two clusters of mutations with largely the same counts of MC3 unique calls (x-axis) and matched variant counts (y-axis). This result led to the decision to keep only remove OxoG artifacts from the overlap criteria. b) MC3 mutations with filter flags also observed in PCAWG data. A Likert1 plot displays the proportion of flagged mutations provided unique to MC3 (blue) and the fraction of mutations present in both MC3 and PCAWG. Total mutations counts are labels below the name of each MC3 filter flag2. Supplementary Figure 4 5 149584 Matching variant type 20331 MC3 Unique 22819 4 PCAWG Unique log10( variant count + 1) Mismatch variant type ## Displays actual counts 3273 1798 3 1393 1150 922 638 2 371 120 18 1 13 10 4 3 3 2 2 2 1 1 1 0 SNP:SNP NA:SNP SNP:NA DEL:DEL SNP:DNP NA:DEL INS:INS DEL:NA NA:INS INS:NA NA:DNP DEL:SNP INS:SNP SNP:TNP SNP:DEL NA:TNP ONP:NA DEL:DNP INS:DEL TNP:TNP ONP:ONP ONP:SNP SNP:INS Variant types (MC3:PCAWG) Supplementary Figure 4: Variant type breakdown. A bar chart shows the breakdown of unique, matched and mismatched variant types in the overlapping dataset. Bars that are purple represent variants with matching variant types in both samples as labeled in mutations annotation formatted files (.maf). Blue bars indicate variants unique to the MC3 data set. Red bars indicate the number of variants unique to the PCAWG data set. Grey bars indicate matched variants with different variant type annotations (x-ax- is). Actual counts are shown at the top of each bar in white. The y-axis is log10 scaled. Supplementary Figure 5 (complement) MuTect/BROAD 35% 15% 50% 9940 (complement) MuSE 38% 39% 23% 11671 (complement) MuSE & MuTect 58% 19% 23% 4464 All Callers Together 11% 77% 12% 202459 MuSE & MuTect 10% 78% 12% 197995 MuSE 9% 79% 12% 190788 MuTect/BROAD 9% 80% 10% 192519 100 50 0 50 100 0 50000 1e+05 150000 2e+05 Percentage n mutations Unique/shared MC3 Shared PCAWG Supplementary Figure 5: Results of extracting subsets of MuSE and MuTect calls and their comple- ments from the full set of mutation calls. Segmented barchart (left) shows that MuSE-only, MuTect-only, and their combined subsets have approximately 80% concordance. Barchart at right shows these 3 subsets each account for most of the calls of the full data set. Conversely, the complementary data-spe- cific call sets (top 3 elements of the segmented chart) exhibit large variations, with concordance as low as 15%, and represent only small fractions of the total number of calls. Overall, differences in software pipelines between WES and WGS do not appear to be a significant confounding factor. Supplementary Figure 6 300 250 Number of samples 200 150 100 50 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Correlation coefficient: mutation signature weights of WES vs WGS Supplementary Figure 6: Assessment of impact on mutation signature analysis. We used SignatureAn- alyzer to predict mutation signatures for WES and WGS data and sought to assess how differences in mutation calls for these data would manifest in higher-level mutational signature analysis. Assessment is shown as a histogram of the Pearson correlation coefficient between the vectors of predicted signa- ture weights from SignatureAnalyzer for matched WGS and WES samples for each of 739 cases. The majority, 555 (75%), show very strong correlation (coefficient at least 90%), with the average correla- tion for the cohort being 89.9%. These samples collectively show a cohort P-value < 2 x 10-6 (Fisher’s transformation). Supplementary Figure 7 a Sources of variation for MC3 unique calls 4136 4000 Intersection Size 3000 2714 2630 2074 2000 1205 1201 1133 1004 1000 742 700 492 438 416 287 203 175 159 144 126 125 109 94 89 87 84 81 80 73 60 60 55 51 50 47 44 43 41 40 39 37 0 GCcontent THCA KICH or PRAD Indels VAF5 MMcomplement Other VarClass VAF10 SubClonal MissensePlus 10000 5000 0 Set Size b Sources of variation for PCAWG unique calls 4774 5000 4674 4000 Intersection Size 2941 3000 2618 2501 2444 2000 1000 753 714 409 405 223 209 195 184 165 164 163 142 136 134 121 104 71 67 56 48 47 42 40 29 29 28 25 25 20 19 19 18 18 15 0 THCA KICH or PRAD VAF5 MMcomplement GCcontent Indels VAF10 Other VarClass MissensePlus SubClonal 15000 10000 5000 0 Set Size Supplementary Figure 7: Sources of variation in private WES calls, and private WGS calls. Two ‘Up- SetR’3 plots were generated to delineate the origin of variation between unique calls in MC3(a) and PCAWG(b). ‘UpSetR’ diagrams provide a unique way to display overlapping sets. Here three variant classifiers sets are used: MissensePlus, Indels, and Other VarClass. An additional six variation source sets are used: GCcontent (GC fraction in 100bp window: 0.7 < GC-content < 0.3) , THCA KICH PRAD (poor performing cancer types), VAF5 (0% < VAF < 5%), VAF10 (5% <= VAF < 10%), SubClonal, MMcomplement (calls originating from callers other than MuSE or MuTect). Supplementary Figure 8 a MC3 Unique to 21,626 PCAWG 15,835 3 W G 2 4,9 7 CA P In MC3 controlled 9,138 b VARSCANI MC3 Filters PINDEL Other common_in_exac INDELOCATOR oxog SOMATICSNIPER broad_PoN_v2 wga VARSCANS nonpreferredpair PASS RADIA MUSE MUTECT 32 64 128 256 512 1024 2048 4096 sqrt(Count) Supplementary Figure 8: Fraction and breakdown of unique PCAWG mutations also found in MC3 con- trolled dataset. a) The relative complements of the MC3/PCAWG Venn diagrams is shown. The right portion of the Venn diagram splits into two sub components in order to illustrate the number of PCAWG variants found in the MC3 controlled dataset. b) The set of 9138 unique PCAWG mutations also de- tected in the MC3 controlled dataset was further separated by MC3 variant caller and MC3 filter flags2. Supplementary Figure 9 A>C A>G A>T C>A C>G C>T Proportion of residuals BLCA BRCA CESC *COAD DLBC GBM HNSC *KICH KIRC KIRP LGG LIHC *LUAD LUSC *OV PRAD READ SARC SKCM STAD THCA Proportion of residuals UCEC 900 300 20 * = Significant FDR < 0.05 ChiSq 100*residuals2 contribution Supplementary Figure 9: Mutation spectrum analysis highlights differences between WGS and WES. Pearson’s chi-squared residuals were calculated to compare the mutations spectrum across cancers from captured exomes versus whole genome sequencing of the same region. The analysis was per- formed by cancer type and the large dot indicates which transition or transversion contributed most to the spectrum difference. The size of the dot is 100-times the Pearson residual for a single transition or transversion. Supplementary References 1. Bryer, J. & Speerschneider, K. likert: Functions to analyze and visualize likert type items. R Packag. version 1, (2013). 2. Ellrott, K. et al. Scalable Open Science Approach for Mutation Calling of Tumor Exomes Using Multiple Genomic Pipelines. Cell Syst. 6, 271-281.e7 (2018). 3. Conway, J. R., Lex, A. & Gehlenborg, N. UpSetR: An R package for the visualization of intersecting sets and their properties. Bioinformatics 33, 2938–2940 (2017).