Algorithm Development for Detection of Low-Prevalence Mutations Through Ultra-Deep Targeted Sequencing

• Activity Number: 668
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lei Bao*+ and Karen Messer
• Companies: University of California at San Diego and University of California at San Diego
• Keywords: Ultra-deep targeted sequencing ; Low prevalence mutation ; Solid tumor ; Statisitcal ranking ; Smoothing ; Sequencing error rate

Abstract

Tumor driver mutations are likely to be present in only a subpopulation of tumor cells. Ultra-deep targeted sequencing (UDT-seq) has emerged as a novel powerful genomic tool to identify low prevalence mutations in solid tumors. We have developed the computational component for the UDT-seq application. Using calibration samples in which we knew the mutation prevalence, we estimated that our method had high sensitivity (94% for prevalence >5%) and specificity (99%). We found that the good performance can be largely attributed to the accurate estimation of the context-dependent sequencing error rate. We estimated a separate error rate stratified by its substitution type, read position and strand, and developed a distance-weighted smoothing technique to smooth the estimates for read positions with low coverage. We also found that although modeling the non-reference allele by a binomial distribution can give a very good ranking of the true positives over false positives, the nominal p-value is overly inflated and cannot be directly used to make any statistical inference. Future work is needed to uncover the exploratory factors that cause the discrepancy and build a better model.