TMB data analysis

The NGS results were analyzed using the default setting of two versions of the integrated pipelines, OTML v1.2 DNA Single Sample on Ion Reporter software v.5.6 and OTML v2.0 DNA Single Sample, available on the Ion Reporter software v.5.10 (Thermo Fisher Scientific, Inc.), which will be referred to as version A and B, respectively.

In both versions of the analysis workflow, the sequenced reads are aligned to human reference genome hg19 and the resulting BAM files are transferred to Ion Reporter software for variant calling. Several parameters of NGS analysis were used to consider a sample suitable for the inclusion in the comparison analysis with the two pipelines: Number of mapped reads, >5,000,000×; mean read depth, >300×; percentage of reads on target, >90%; target base coverage at 100×, >95%; uniformity, >80%. TMB calculation does not require matched normal samples. After the variant calling, a filter chain was applied to remove germline variants using population databases (1000 Genome Project, https://www.internationalgenome.org/data/; NHLBI GO Exome Sequencing Project https://esp.gs.washington.edu/drupal/node/1; and ExAC, http://bigd.big.ac.cn/databasecommons/database/id/3774), and to select somatic variants which have a minimum depth of base coverage above 60× to be used for TMB calculation without generating noise. In version A, TMB is calculated by counting synonymous and non-synonymous somatic single-base substitutions (SNVs), at or above 5% allelic frequency (AF), from the full panel, which is 1.2 Mb exonic and 0.45 Mb intronic. In version B, only the non-synonymous SNVs and short insertion-deletion mutations (InDels) with ≥5% AF are considered, derived from the 1.2-Mb exonic region. The number of mutations counted is divided by the number of bases with sufficient coverage, to normalize the TMB values. The variant calling in version A is optimized only for the TMB calculation and not for variant detection, whereas the workflow in version B is also able to call clinically relevant variants in a specific predefined set of genes with a limit of detection (LOD) of 5% AF. Finally, following application of the filter chain, workflow B applies calibration to bring the TMB values closer to WES-based TMB. The calibration factor was calculated through linear modelling on The Cancer Genome Atlas database (https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga). Workflow A does not apply calibration.

At the end of each analysis, a report was provided that included the normalized mutation load, defined as the number of somatic mutations per Mb of genome with an average coverage not less than 300×. In addition, the report summarized other information on the samples, including coverage, variants called, mutation signatures of the somatic variants, mutations consistent with deamination, UV and tobacco smoke damage.