neoepiscopeimproves neoepitope prediction with multi-variant phasing

Abstract

The vast majority of tools for neoepitope prediction from DNA sequencing of complementary tumor and normal patient samples do not consider germline context or the potential for co-occurrence of two or more somatic variants on the same mRNA transcript. Without consideration of these phenomena, existing approaches are likely to produce both false positive and false negative results, resulting in an inaccurate and incomplete picture of the cancer neoepitope landscape. We developed neoepiscope chiefly to address this issue for single nucleotide variants (SNVs) and insertions/deletions (indels), and herein illustrate how germline and somatic variant phasing affects neoepitope prediction across multiple datasets. We estimate that up to ~5% of neoepitopes arising from SNVs and indels may require variant phasing for their accurate assessment. neoepiscope is performant, flexible, and supports several major histocompatibility complex binding affinity prediction tools. We have released neoepiscope as open-source software (MIT license, https://github.com/pdxgx/neoepiscope ) for broad use. KEY POINTS• Germline context and somatic variant phasing are important for neoepitope prediction • Many popular neoepitope prediction tools have issues of performance and reproducibility • We describe and provide performant software for accurate neoepitope prediction from DNA-seq data

Burden of tumor mutations, neoepitopes, and other variants are weak predictors of cancer immunotherapy response and overall survival

“…Despite attempts to obtain these data, we unfortunately were forced to omit tumor samples from 75 NSCLC patients [19], for whom data was not available due to limitations of patient consent at the time of the study. Alignment of whole exome sequencing (WES) reads was performed as described previously [20]. The Mbp of genome covered was determined using bedtools genomecov (v2.26.0) [21], where any base covered by a depth of at least six reads was considered covered, as this is twice the minimum read depth required for variant detection by Soma-ticSniper [22] and VarScan 2 [23].…”

“…The Mbp of genome covered was determined using bedtools genomecov (v2.26.0) [21], where any base covered by a depth of at least six reads was considered covered, as this is twice the minimum read depth required for variant detection by Soma-ticSniper [22] and VarScan 2 [23]. Somatic and germline variant calling were performed as described previously [20]. To obtain coverage-adjusted mutation burdens for each patient, we divided the number of consensus somatic variants by the Mbp of genome covered by sequencing.…”

“…We employed HapCUT2 for patient-specific haplotype phasing. To do this, germline and consensus somatic variants were combined into a single VCF using neoepiscope's [20] (v0.3.5) merge functionality. HapCUT2's extractHAIRS software was run with the merged VCF and the tumor alignment file, allowing for extraction of reads spanning indels, to produce the fragment file used with HapCUT2 to predict haplotypes.…”

Burden of tumor mutations, neoepitopes, and other variants are dubious predictors of cancer immunotherapy response and overall survival

“…due to limitations of patient consent at the time of the study. Alignment of whole exome sequencing (WES) reads was performed as described previously (23) .The Mbp of genome covered was determined using bedtools genomecov (v2. 26.0) (24) , where any base covered by a depth of at least 6 reads was considered covered, as this is twice the minimum read depth required for variant detection by SomaticSniper (25) and VarScan 2 (26) .…”

“…26.0) (24) , where any base covered by a depth of at least 6 reads was considered covered, as this is twice the minimum read depth required for variant detection by SomaticSniper (25) and VarScan 2 (26) . Somatic and germline variant calling were performed as described previously (23) . To obtain coverage-adjusted mutation burdens for each patient, we divided the number of consensus somatic variants by the Mbp of genome covered by sequencing.…”