Pindel-TD: A Tandem Duplication Detector Based on A Pattern Growth Approach

Abstract

Tandem duplication (TD) is a major type of structural variations (SVs) that plays an important role in novel gene formation and human diseases. However, TDs are often missed or incorrectly classified as insertions by most modern SV detection methods due to the lack of specialized operation on TD-related mutational signals. Herein, we developed a TD detection module for the Pindel tool, referred to as Pindel-TD, based on a TD-specific pattern growth approach. Pindel-TD is capable of detecting TDs with a wide size range at single nucleotide resolution. Using simulated and real read data from HG002, we demonstrated that Pindel-TD outperforms other leading methods in terms of precision, recall, F1-score, and robustness. Furthermore, by applying Pindel-TD to data generated from the K562 cancer cell line, we identified a TD located at the seventh exon of SAGE1, providing an explanation for its high expression. Pindel-TD is available for non-commercial use at https://github.com/xjtu-omics/pindel.

1. Introduction

Tandem duplication (TD) is one of the major types of structural variation (SV) [1,2], contributing to the formation of de novo gene structure [3], the evolution of biosynthetic pathways in plants [4], and various human diseases, including autism [5] and cancers. Cancers such as breast, ovarian, and endometrial carcinomas can be further divided into different tandem duplicator phenotype (TDP) subgroups based on the frequency and distribution of TDs [6,7]. TDs contribute to tumorigenesis by augmenting oncogene expression and disrupting tumor suppressor genes [6,7].

Fueled by the development of next-generation and third-generation sequencing technologies, numerous SV detection methods have been developed based on model-matched or model-free strategies. In general, model-matched strategies involve extracting the mutational signal, such as discordant read-pairs, read-depth, and split-reads, and then matching the signal to a specific SV model. For example, Pindel utilizes a pattern growth approach to leverage split-read signals for precise breakpoint SV detection [8,9], while DELLY performs an integration analysis of the pair-end mapping and split-read information to detect SVs at single nucleotide resolution [10]. LUMPY was developed to precisely identify SVs by integrating three different mutational signals of read-pair, split-read, and read-depth using general probabilistic framework [11]. While model-matched SV detection strategies are efficient in detecting simple SVs, e.g., insertion, deletion, and inversion, they often lack performance for complex SVs (CSVs) that contain multiple breakpoints and play important roles in cancer [9,12]. Mako [13] and SVision [14] are two model-free methods designed to detect CSVs from the short-read and long-read sequencing data, respectively. Although these methods perform well for the most frequent types of SVs, such as insertions and deletions, they exhibit lower accuracy and recall rates for certain types of SVs, such as TDs, which are frequently reported as insertions [15,16] due to the lack of specific optimization on TDs. Therefore, developing a TD detection tool specifically optimized to accurately characterize TDs is an eminent need in the genomic community.