Dragen 3

Insert size distributions across the entire genome were calculated using Illumina Dragen 3.7.3. The ratio of the number of fragments with sizes less than or equal to 150 bp and greater than 150 bp but less than 500 bp were calculated for all samples. Tumor fraction was estimated for all samples using the ichorCNA variant calling pipeline as previously described in ref. ^{38 (link)}. Briefly, the algorithm uses a Hidden Markov Model (HMM) to predict copy number segments and to estimate the circulating tumor content from total cfDNA sequenced. The details of this approach have been previously described in ref. ^{38 (link)}. For each of the categories (i.e., proportion of fragments <150 bp, tumor fraction, and tumor concentration) samples were classified by tumor type, and a Kruskal–Wallis test was performed between cancer types. Only samples collected at the time of diagnosis were used for this analysis. A p value <0.01 was considered statistically significant.

Reads were aligned to the 1000 genomes phase 2 reference genome (hs37d5) which includes build GRCh37 and decoy sequences: (ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/technical/reference/phase2_reference_assembly_sequence/hs37d5.fa.gz) using the Illumina Dragen 3.7.3 aligner. The ichorCNA algorithm was subsequently applied using the depth of coverage obtained from 500 kb bins across the genome^{38 (link)}. When available, LP-WGS data from the LB samples were compared with CNA profiles of the matching primary tumors generated from CMA assays (CytoScanHD or OncoScan, Thermo Fisher Scientific, Waltham, MA). Briefly, all 59 Cytoscan or OncoScan arrays performed on the tumor tissue were processed using ASCAT and NxClinical (BioDiscovery, El Segundo, CA) using the default settings, except that the Piecewise Constant Fitting (PCF) penalty was increased to 95 to account for FFPE degradation of samples. Subsequently, the segmented calls from LP-WGS and CMA were binned into 1 Mb regions across the entire genome. Pearson’s correlation for each pair of samples across all bins over autosomes was calculated for all CMA samples and LP-WGS samples with purity greater than 10%. All samples with a Pearson’s correlation coefficient (r²) less than 0.7 were manually examined.