Lightning-fast genome variant detection with GROM

Abstract

Current human whole genome sequencing projects produce massive amounts of data, often creating significant computational challenges. Different approaches have been developed for each type of genome variant and method of its detection, necessitating users to run multiple algorithms to find variants. We present Genome Rearrangement OmniMapper (GROM), a novel comprehensive variant detection algorithm accepting aligned read files as input and finding SNVs, indels, structural variants (SVs), and copy number variants (CNVs). We show that GROM outperforms state-of-the-art methods on 7 validated benchmarks using 2 whole genome sequencing (WGS) data sets. Additionally, GROM boasts lightning-fast run times, analyzing a 50× WGS human data set (NA12878) on commonly available computer hardware in 11 minutes, more than an order of magnitude (up to 72 times) faster than tools detecting a similar range of variants. Addressing the needs of big data analysis, GROM combines in 1 algorithm SNV, indel, SV, and CNV detection, providing superior speed, sensitivity, and precision. GROM is also able to detect CNVs, SNVs, and indels in non-paired-read WGS libraries, as well as SNVs and indels in whole exome or RNA sequencing data sets.

Structural variants in 3000 rice genomes

“…Structural variants can be classified in the following types: deletions, insertions, duplications (tandem and interspersed), inversions, and translocations. There are five general strategies to detect SVs based on analysis of data from high-throughput sequencing data using short reads: paired-end mapping (RP) (Chen et al 2009;Sindi et al 2009), split-read mapping (SR) (Schröder et al 2014), read depth (RD) (Abyzov et al 2011;Duitama et al 2014;Smith et al 2015), de novo assembly (AS) (Narzisi et al 2014;Rizk et al 2014;Yang et al 2015), and a combination of the preceding approaches (CB) (Ye et al 2009;Rausch et al 2012;Layer et al 2014;Mohiyuddin et al 2015;Smith et al 2017a). Each of these strategies has different strengths and weaknesses in detection, depending on variant type, sequence length, and reference genome quality and complexity; hence, applying complementary methods and combining results can overcome some of the limitations inherent to these different approaches (Alkan et al 2011).…”

ARIADNA: machine learning method for ancient DNA variant discovery

“…A combination of GROM and ARIADNA is also much faster than GATK and AntCaller (snpAD has not been released and was not available for testing). In a direct comparison, 33 GROM was 12–25 times faster than GATK on a single thread and more than 70 times faster on 24 threads. In our tests, AntCaller was 10–20 times slower than GROM on a single thread, somewhat faster than GATK, as in earlier AntCaller comparisons with GATK.…”

“…Through a series of these trees, prediction deviance from known truth is continually reduced. Our feature set is generated using a modification of our comprehensive mutation caller GROM 33 to act as a genome scanner and output feature information at potential mutation locations. These features include common measures such as read depth, SNV count, read and base quality as well as features unique to aDNA, such as distance from read end, C→T substitutions, and neighbouring mutation rates ( Table 1 ).…”

“…Given the lack of validated ground truth datasets for aDNA, it uses common and unique variants in multiple woolly mammoth genomes to train a predictive machine learning (ML) model. ARIADNA employs our fast GROM genome scanning engine 33 to find all potential SNVs (PSNVs) found as deviations between sample and reference genomes and then utilizes a boosted regression tree algorithm for training and classification of potential mutation sites. The unique features of the corresponding sites are used by our algorithm to determine the difference between bona fide mutations in aDNA and noise due to aDNA degradation or contamination.…”

Clinical sequencing: From raw data to diagnosis with lifetime value

“…CNVs and copy neutral rearrangements such as inversions and translocations are SVs affecting large genomic segments . Except for the software GROM (Genome Rearrangement OmniMapper), a recently introduced all‐in‐one solution to detect SNVs, indels, CNVs, and other SVs, algorithms used for calling of SNVs and indels are not suited for the detection of larger sequence variants, requiring dedicated algorithms for CNV detection. Calling of CNVs from HTS data can be achieved by different approaches such as read‐depth analysis, paired‐end mapping, split reads, and de novo assembly .…”