GROM-RD: resolving genomic biases to improve read depth detection of copy number variants.

Sean D Smith,Andrey Grigoriev,Joseph K Kawash

doi:10.7717/peerj.836

Abstract

Amplifications or deletions of genome segments, known as copy number variants (CNVs), have been associated with many diseases. Read depth analysis of next-generation sequencing (NGS) is an essential method of detecting CNVs. However, genome read coverage is frequently distorted by various biases of NGS platforms, which reduce predictive capabilities of existing approaches. Additionally, the use of read depth tools has been somewhat hindered by imprecise breakpoint identification. We developed GROM-RD, an algorithm that analyzes multiple biases in read coverage to detect CNVs in NGS data. We found non-uniform variance across distinct GC regions after using existing GC bias correction methods and developed a novel approach to normalize such variance. Although complex and repetitive genome segments complicate CNV detection, GROM-RD adjusts for repeat bias and uses a two-pipeline masking approach to detect CNVs in complex and repetitive segments while improving sensitivity in less complicated regions. To overcome a typical weakness of RD methods, GROM-RD employs a CNV search using size-varying overlapping windows to improve breakpoint resolution. We compared our method to two widely used programs based on read depth methods, CNVnator and RDXplorer, and observed improved CNV detection and breakpoint accuracy for GROM-RD. GROM-RD is available at http://grigoriev.rutgers.edu/software/.

Highlights

Copy number variants (CNVs) have been linked to several diseases including cancer (Berger et al, 2011; Campbell et al, 2010; Stephens et al, 2009), schizophrenia (Stefansson et al, 2009), and autism (Marshall et al, 2008)
We have developed GROM-Read depth (RD), a control-free whole genome sequencing (WGS) RD algorithm with several improvements and novel features compared to existing RD algorithms, such as excessive coverage masking, GC bias mean and variance normalization, GC weighting, dinucleotide repeat bias detection and adjustment, and a size-varying sliding window CNV search
RSVSim assumed a beta distribution to create a distribution of CNV sizes based on SVs from the Database of Genomic Variants with lengths between 500 and 10,000 bases, resulting in a decreasing frequency of CNVs with increasing size

Summary

INTRODUCTION

Copy number variants (CNVs) have been linked to several diseases including cancer (Berger et al, 2011; Campbell et al, 2010; Stephens et al, 2009), schizophrenia (Stefansson et al, 2009), and autism (Marshall et al, 2008). We have developed GROM-RD, a control-free WGS RD algorithm with several improvements and novel features compared to existing RD algorithms, such as excessive coverage masking, GC bias mean and variance normalization, GC weighting, dinucleotide repeat bias detection and adjustment, and a size-varying sliding window CNV search. These features address weaknesses in existing RD methods and biases in genomic sequencing that limit CNV sensitivity, specificity, and breakpoint accuracy, as evidenced by comparison of our algorithm to two most commonly used control-free WGS RD tools, RDXplorer (Yoon et al, 2009) and CNVnator (Abyzov et al, 2011). GROM-RD showed improved predictive capabilities and breakpoint resolution for CNVs, as well as excellent scalability for different NGS datasets, both simulated and real

METHODS

RESULTS

DISCUSSION