Dilution series of two cancer cell lines analyzed using ultra deep duplex sequencing.

Detection of low-frequency mutations constitutes a critical challenge in oncology and precision medicine. As next-generation sequencing and liquid biopsy become more prevalent in the clinic and research, there is an increasing need to improve the sensitivity of mutation detection as well as other sources of cancer genome variation. Technical artefacts are introduced by sequencing technologies and a common solution utilizes unique molecular identifiers (UMIs) to distinguish errors from true variants found across multiple reads of the same molecule. While these methods are effective, they often rely on over-amplification and redundant sequencing resulting in inefficient error correction at lower sequencing depths. Here, we describe ConsensusCruncher, an open-source python software program for generalized processing of UMI-tagged, paired-end sequencing reads that enables double-strand error suppression for both redundant (consensus sequences) and non-redundant (singleton) reads. Traditional UMI-based error suppression methods only correct redundant reads and have strict UMI specifications. ConsensusCruncher supports UMIs of any length and produces versatile consensus sequences in standardized binary alignment map (BAM) files that are primed for downstream calling of mutations, copy number alterations, and structural rearrangements. We provide best practices for variant calling and quantification, using consensus sequences for highly accurate detection and all unique reads for quantification.

低频突变检测是肿瘤学与精准医学领域的一项关键挑战。随着下一代测序（next-generation sequencing）与液体活检（liquid biopsy）在临床与科研领域的应用愈发普及，提升突变检测及其他癌症基因组变异类型检测灵敏度的需求日益增长。测序技术会引入技术伪影，目前主流解决方案借助唯一分子标识符（unique molecular identifiers, UMIs），将测序错误与同一分子多次测序读段中的真实变异区分开来。尽管此类方法具备一定有效性，但通常依赖过度扩增与冗余测序，导致低测序深度下的错误校正效率低下。本研究介绍了ConsensusCruncher——一款开源Python软件，可通用处理带UMI标记的双端测序读段，能够对冗余（共识序列）与非冗余（单例）读段均实现双链错误抑制。传统的基于UMI的错误抑制方法仅能校正冗余读段，且对UMI有着严格的规格要求。ConsensusCruncher支持任意长度的UMI，可在标准化二进制比对映射（binary alignment map, BAM）文件中生成多样化的共识序列，可为下游的突变检测、拷贝数变异与结构重排分析做好准备。本研究还提供了变异检测与定量分析的最佳实践方案：借助共识序列实现高精度变异检测，利用全部唯一读段完成定量分析。