Abstract 4850: Identifying copy number alterations from targeted sequencing data

Abstract

Abstract Identification of copy number alterations (CNAs) in tumors can assist in determining diagnosis, prognosis, or predicting response to therapy in certain cancer types. Many CNA detection tools have been developed for whole genome sequencing (WGS). These algorithms typically “call” alterations by using significant deviations of the log2 ratio of tumor to normal coverage from the expected diploid log2 ratio. Several factors, however, including reporting pertinent genomic information, hardware and software limitations, and cost, have impeded the routine application of WGS in a clinical setting. Targeted sequencing - hybrid capture and massively parallel sequencing of defined regions (targets) of the genome can overcome some of these challenges and can specifically interrogate, for example, known cancer genes, thus maximizing the “actionability” or utility of the information generated. Though more suited for large-scale clinical testing, targeted sequencing of smaller gene panels with specific clinical reference presents some significant bioinformatic challenges that hinder the application of pre-existing CNA detection tools, resulting in a heavy dependence on manual review in the clinic. In addition, varying quality of DNA, minor variations in laboratory protocols, along with batch effects, are known to confound CNA profiles and are exasperated by the decreased number of sequenced regions of targeted gene panels. These variables alter the log2 ratio, and thus, our ability to call CNAs suffers. We note, that using matched normals may mitigate some deleterious effects, however, to minimize clinical cost and throughput, we tackle these challenges with unmatched tumor-normal pairs. Similar to the closest normal approach used in SNP array analyses, we rely on the assumption that some “normal” (non-tumor) samples experience similar experimental variability that affects the tumor samples. Minimizing maximally informative cost-functions (incorporating both global and local CNA variation), we iteratively investigate different combinations of normal samples which best resemble a given cancer sample. The new set of normals which minimize the cost-function, defines a panel of normals (PON) whose target medians are used to generate the log2 ratio. Along with batch-effect reduction techniques, we then call CNAs with statistically significant thresholds at the target-level. Interestingly, we find that approximately 15 normals are typically needed to minimize CNA profile variation, as opposed to our typical approach of using all available normals (∼60). In addition, we demonstrate that our procedure can improve our sensitivity and specificity from 58±1% to 81±2% and 94±1% to 99.5±0.1%, respectively, on simulated data. We further compared our approach with other published algorithms and found a significant improvement in performance, reducing the time required to manually review CNAs and interpret copy number profiles in the clinic. Citation Format: Bernard Fendler, Ryan Abo, Samuel Hunter, Matthew Ducar, Elizabeth Garcia, Paul Van Hummelen, Neal Lindeman, Laura MacConaill. Identifying copy number alterations from targeted sequencing data. [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 4850. doi:10.1158/1538-7445.AM2015-4850