On Statistical Modeling of Sequencing Noise in High Depth Data to Assess Tumor Evolution

Abstract

One cause of cancer mortality is tumor evolution to therapy-resistant disease. First line therapy often targets the dominant clone, and drug resistance can emerge from preexisting clones that gain fitness through therapy-induced natural selection. Such mutations may be identified using targeted sequencing assays by analysis of noise in high-depth data. Here, we develop a comprehensive, unbiased model for sequencing error background. We find that noise in sufficiently deep DNA sequencing data can be approximated by aggregating negative binomial distributions. Mutations with frequencies above noise may have prognostic value. We evaluate our model with simulated exponentially expanded populations as well as data from cell line and patient sample dilution experiments, demonstrating its utility in prognosticating tumor progression. Our results may have the potential to identify significant mutations that can cause recurrence. These results are relevant in the pre-treatment clinical setting to determine appropriate therapy and prepare for potential recurrence pretreatment.

Association of JAK2-V617F Mutations Detected by Solid Tumor Sequencing With Coexistent Myeloproliferative Neoplasms

“…To validate that JAK2 mutations existed in hematopoietic elements, we performed manual macrodissection on paraffin-embedded specimens to enrich for cancer or hematopoietic cells and sequenced these samples at high depth (>2500×) using a 49-gene panel (RainDance Technologies) on Illumina MiSeq. We identified all sites that were different from the reference using an inclusive variant caller and used a Bayesian approach to detect true mutations against background error in which mutations were tested in each sample against 33 previously sequenced, JAK2 wild-type samples. After correcting for multiple hypotheses using the Benjamini-Hochberg method, we generated a list of variants with a false-discovery rate less than 0.001.…”

Subclonal NT5C2 mutations are associated with poor outcomes after relapse of pediatric acute lymphoblastic leukemia

“…To detect true mutations against background error, we used a Bayesian approach as described previously. 28 Because of variation in sequencing depth and mutation-specific sensitivity across the gene and between samples, we applied an uniform 2% variant allele frequency threshold for mutation calling. Details are given in the supplemental Methods.…”

A Bayesian Nonparametric Model for Inferring Subclonal Populations from Structured DNA Sequencing Data

“…We note that previous work has shown that DNA sequencing count data is well-represented by a negative binomial distribution (Rabadan et al, 2018).…”