Abstract 4892: Methods for accurate reporting of confidence intervals in clinical applications of next generation sequencing (NGS)

Abstract

Abstract BACKGROUND: A challenge to clinical implementation of Next-Generation Sequencing (NGS) is lack of appropriate quality control including control for: a) adequate sample loading, b) variation in target amplification efficiency, and c) variation in loading of prepared NGS specimen onto sequencing platform. Polymerase chain reaction steps used in preparation for NGS can yield a large number of sequencing reads from a small number of starting nucleic acid molecules (e.g. small and/or degraded samples from FNA specimens or FFPE tissue), resulting in large stochastic sampling variation. At present, methods to quantify the analyzable fraction of target nucleic acid are not available or are not suitable for small specimens. As a result, current practice is to rely on sequencing coverage data that may provide false assurance of adequate specimen sampling during molecular analysis, and this has potential to negatively impact patient care. We hypothesize that coefficient of variation (CV) for amplicon-based NGS assay measurements is largely predicted by Poisson (i.e. stochastic) sampling effects for a nucleic acid target at two key points: 1) input molarity (i.e. number of intact molecules) and 2) sequencing coverage (i.e. read counts). METHODS: To test this hypothesis we developed three working models using Monte Carlo simulation and derived equations to predict expected CV based on sequence read count and/or intact molecules mesaured for a given nucleic acid target. These expectation models were tested against empirically derived data from cross-mixtures of two cell lines (H23 and H520) known to be homozygous for opposite alleles at four polymorphic sites (rs769217, rs1042522, rs735482 and rs2298881). Cell lines were mixed to produce limiting inputs of one allele relative to the other, then prepared for NGS such that a broad range of combinations of limiting allelic molecule inputs and/or sequence read counts were observed (46 sets of allelic measurement at all 4 loci). Intra-assay measurement of intact and amplifiable molecules was accomplished using recently described competitive multiplex-PCR amplicon-based NGS specimen preparation (Blomquist, et. al. 2013). RESULTS: Observed CV for measurement at varying input copies and sequencing read counts were compared to expected CV. Actual measured CV was on average 13- and 1.5-fold higher than expected CV based on sequencing reads or molecule input measurements alone, respectively. For the model derived from both sequencing reads and molecule input measurements, expected CV was very close to measured (average [measured CV/expected CV] = 1.01) and explained 74% of observed assay variance. CONCLUSIONS: NGS-based diagnostic tests that do not take into account both input concentration of intact target nucleic acid material and associated sequencing read coverage may not provide accurate reporting of confidence intervals in specimens with limited or degraded material. Citation Format: Erin L. Crawford, Thomas Blomquist, James C. Willey. Methods for accurate reporting of confidence intervals in clinical applications of next generation sequencing (NGS). [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 4892. doi:10.1158/1538-7445.AM2015-4892