The ICR639 CPG NGS validation series: A resource to assess analytical sensitivity of cancer predisposition gene testing

Abstract

The analytical sensitivity of a next generation sequencing (NGS) test reflects the ability of the test to detect real sequence variation. The evaluation of analytical sensitivity relies on the availability of gold-standard, validated, benchmarking datasets. For NGS analysis the availability of suitable datasets has been limited. Most laboratories undertake small scale evaluations using in-house data, and/or rely on in silico generated datasets to evaluate the performance of NGS variant detection pipelines. Cancer predisposition genes (CPGs), such as BRCA1 and BRCA2, are amongst the most widely tested genes in clinical practice today. Hundreds of providers across the world are now offering CPG testing using NGS methods. Validating and comparing the analytical sensitivity of CPG tests has proved difficult, due to the absence of comprehensive, orthogonally validated, benchmarking datasets of CPG pathogenic variants. To address this we present the ICR639 CPG NGS validation series. This dataset comprises data from 639 individuals. Each individual has sequencing data generated using the TruSight Cancer Panel (TSCP), a targeted NGS assay for the analysis of CPGs, together with orthogonally generated data showing the presence of at least one CPG pathogenic variant per individual. The set consists of 645 pathogenic variants in total. There is strong representation of the most challenging types of variants to detect, with 339 indels, including 16 complex indels and 24 with length greater than five base pairs and 74 exon copy number variations (CNVs) including 23 single exon CNVs. The series includes pathogenic variants in 31 CPGs, including 502 pathogenic variants in BRCA1 or BRCA2, making this an important comprehensive validation dataset for providers of BRCA1 and BRCA2 NGS testing. We have deposited the TSCP FASTQ files of the ICR639 series in the European Genome-phenome Archive (EGA) under accession number EGAD00001004134.

Assessment of structural chromosomal instability phenotypes as biomarkers of carboplatin response in triple negative breast cancer: the TNT trial

“…Pathogenic germline variants in DDR genes 18 increase the risk of developing cancer and were identified in peripheral blood leukocyte DNA in 8/135 patients, not including BRCA1/2 ( supplementary Table S5 , available at https://doi.org/10.1016/j.annonc.2020.10.475 ). The majority (62.5%, 5/8) of these cases were part of the low HLAMP group and were completely absent in the high group ( Figure 2 B).…”

“…Illumina TruSight Cancer v2 targeted sequencing panel 18 was used to identify pathogenic germline variants of 97 genes associated with predisposition to cancer.…”

A machine learning approach based on ACMG/AMP guidelines for genomic variant classification and prioritization

“…Finally, ML performances were tested on real data from 639 individuals that underwent NGS test to detect pathogenic variants in Cancer predisposition genes, such as BRCA1 and BRCA2 2 . VCF files for each donor, along with the list of pathogenic variants reported in this cohort, have been made available to the research community in 2018 2 .…”

“…Thanks to next generation sequencing (NGS) technologies, thousands of variants can be identified for a single patient depending on the type of experimental assay (e.g., gene panel, whole exome or genome). In the context of monogenic inherited disorders, only few of these variants might be pathogenic for the condition of interest 2 . In order to identify them, clinical laboratory geneticists often leverage on in silico functional tools that are able to assess protein variant intolerance, splicing or gene regulatory alterations 3 .…”

“…the ability to correctly discriminate pathogenic and benign variants) and prioritization (i.e. the ability to prioritize true pathogenic variants) on real variant data 2 . Classification and prioritization of the proposed ML approaches were compared with classification and prioritization performed by guidelines-based scoring methodologies (the pathogenicity score suggested in Nicora et al 19 and the Bayesian modeling suggested in Tavtigian et al 21 ) and with two widely used in silico scoring methods, VVP 31 and CADD 32 .…”