Abstract 5093: Clinical-grade NGS analytic platform with explicit negative calling to improve variant reporting across different laboratory protocols: Accuracy study

Abstract

Abstract Introduction: Research-grade bioinformatic pipelines share a common workflow with a logical shortcut that infers where no variant was called, no variant exists. The complexity of NGS protocols makes this inference clinically unsound. Incomplete enrichment, or a sample with deletions in the region of interest can result in false negative errors. Clinical-grade NGS analytics should explicitly evaluate each region of clinical interest to distinguish regions with no informative variants from those with too little evidence to make a call. An ideal platform would produce reliable results from sequencers and library prep in clinical use without protocol-specific tuning. This study evaluates the performance of a novel Clinical-grade Analytic Platform designed to meet these challenges. Study Design: 189 samples from the NA12878 cell line, processed by 13 different laboratories, using 18 different protocols and 11 different instruments were sourced as raw sequencing data from public repositories. Each sample file was run through the Clinical-grade Analytic Platform to produce a cVCF result, a detailed accounting of variants present, regions with no variant present and regions with insufficient coverage across 14,626 exons in the 1043-gene PanCancer TestPanel. High Confidence TRUTH, obtained from the GIAB consortium, established that NA12878 DNA contains 598 variants and 14,230 exons with no variant across the 1043 genes. The cVCFs were compared to TRUTH to quantify performance. Results: The Clinical-Grade NGS Analytic Platform ran all 189 samples without protocol or laboratory-specific tuning, with an overall sensitivity of 95% and specificity of 99%. This performance includes samples with average read depth ranging from 18x to 10,000x, although regions with local read depths above 30x performed better. Hybridization capture-based methods generally outperformed amplicon-based methods at all read depths. Full presentation will discuss performance stratified by instrument, library prep method, local read depth, and variant type. Conclusion: The Clinical-grade Analytic Platform, Farsight Correo, showed strong performance across a broad range of read depths, instruments and library prep methods. The platform’s unique ability to distinguish between “No Variant Present” and “Insufficient Coverage” contributed significantly to this performance. Robust, cross-protocol performance is clinically valuable. It can enable clinical laboratories to expand or change NGS protocols without replacing analytics. It can enable companion diagnostic programs to include multiple sequencing protocols and partners. And finally, explicit negative variant calling can help AI and machine learning mirror the underlying biology; integrating variant presence and absence into new pathway-based models to significantly advance genomics-guided medicine. Citation Format: Edward P. Tang, Charlie C. Kim, Kristine C. Mechem, Glenda G. Anderson. Clinical-grade NGS analytic platform with explicit negative calling to improve variant reporting across different laboratory protocols: Accuracy study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 5093.