Dr Joanna Przybyl provides an approach for profiling molecular changes in leiomyosarcoma (LMS) tumor-derived DNA using targeted sequencing and copy number analysis, in order to detect molecular alterations that could provide clues as to potential therapeutic targets for LMS tumors.
DRAGEN v4.2 integrates depth signals from CNV callers with junction signals from SV callers to increase its sensitivity in detecting small to medium genomic gain or loss events.
Detection of Copy Number Variants
Copy number variants (CNVs) are structural variants that disrupt DNA copying and genome regulation, often having significant biological ramifications such as altered gene expression. CNVs differ from SNPs and Indels in that they occupy larger genomic space ranging from several thousand bases up to 5 megabases [1].
Structural variation analysis has traditionally been accomplished via array comparative genomic hybridization (aCGH). Next-generation sequencing (NGS), however, allows CNV detection with higher resolution than aCGH and may enable identification of CNVs that had previously gone undetected due to other methods of detection. NGS’ high resolution capability makes this technology particularly valuable when searching for CNVs that alter DNA replication or transcription – potentially implicated in cancer or other disorders.
NGS may provide numerous advantages over A-CGH; however, detecting CNVs can still be challenging due to an increased number of data points and more complex nature of NGS data sets. Furthermore, NGS methods may be vulnerable to short read lengths or bias in GC content biasing, thus making aCGH as a quality control measure crucial in order to reliably detect CNVs in NGS data sets.
We conducted a comprehensive analysis on 67 samples from six global buffalo breeds using aCGH and WGS data to detect CNVs per sample and per CNV, after quality control on an individual and per CNV basis. Following quality checks we identified 471 GEN_CNV variants and 1,855 WGS_CNV variants as potential CNVs.
On average, WGS_CNV were detected with higher experiment-wide genome coverage than GEN_CNV due to strict CNVR discovery parameters and small sample size limitations. Conversely, GEN_CNV were present across many genomic regions with similar length distributions as aCGH experiments.
Genotyping signal intensity information to detect CNVs is a widely adopted practice, and several in silico methods have been developed for this purpose. However, experimental validation with qPCR or FISH remains necessary in order to provide true validation; to expedite this process more efficiently an accurate computational method must also be available; herein we have implemented an algorithm based on a Bayesian information criterion (BIC) algorithm which detects CNVs with high confidence from genomic data sets.
Detection of Deletions
Copy number variants (CNVs), especially deletions, is an essential step toward understanding how genomic variations impact phenotype. A variety of techniques have been employed to detect CNVs including comparative genomic hybridisation arrays, single nucleotide polymorphism (SNP) arrays and depth-of-coverage metrics applied to whole genome sequence data.
CNV detection approaches depend heavily on the quality of original source data. Combinations of methods may help increase sensitivity while simultaneously decreasing false positives generated by each technique individually.
To assess the performance of different CNV detection methods, we have conducted an artificially constructed dataset and evaluated their performance on it. This dataset includes randomly sampled genomic positions of autosomal CNVs and chromosome X duplications with pairwise distance measurements between them; then a Kolmogorov-Smirnov test was run to establish whether their distribution differed significantly from that expected if all CNVs had been evenly spread throughout the genome.
We compared the performance of various genomic CNVR detection methods by analyzing sequencing data from one animal. Our results indicate that combining CNV detection methods may allow for increased deletion detection compared to individual methods alone; however, please be aware that individual detection methods differ considerably in how much deletions they detect.
BIC-seq, a nonparametric algorithm developed for genomic data, was created to detect CNVs using probabilities that each read belonged to one of four mixture models: normal diploid copy number (CN=0), homozygous deletion (CN=-1), heterozygous duplication (2-3), and heterozygous duplication (2-3). Model parameters were estimated using logistic regression. BIC-seq’s robustness against outliers makes it ideal for large datasets generated from next-generation sequencers capable of up to 30x coverage coverage.
We applied BIC-seq to a tumor genome sequence from a patient with glioblastoma and its matching normal sample from another individual, successfully detecting somatic CNVs in 22 genes including cancer-related genes EGFR and CDKN2A that show both amplified copies and deletions that frequently appear in these tumors.
Detection of Duplications
CNVs differ from SNPs in that they alter an entire genomic region’s amount of DNA present; such changes can have various impacts, including altering protein expression levels, exposing recessive alleles and gene duplication/deletion. Because CNVs play such an integral part of livestock genetic improvement programs, their identification has become essential. Multiple methods have been devised for CNV identification including comparative genomic hybridization arrays (CGH) arrays and SNP genotyping arrays.
Contrast SNP arrays with CGH arrays which report relative signal intensities. SNP arrays collect normalized total signals and allelic intensity ratios representing overall copy number and allele frequency; however, data collected using SNP arrays is usually limited in size due to how they’re collected, which could introduce bias into their CNV detection methods. Furthermore, most SNP arrays were created for genome-wide association studies without including probes specifically optimized for CNV detection.
Genomic CNV wave whole genome sequencing offers an alternative approach for detecting CNVs with higher sensitivity and specificity than SNP arrays. To facilitate detection, software used in this study partitioned the genome into non-overlapping bins that were assessed based on read depth comparison with an applicable threshold; additionally, statistical tests were run against CNV calls to validate accuracy of calls.
Genomic cnv wave software was put through rigorous accuracy tests using data generated by the Comparative Genomics Laboratory at University of Wisconsin-Madison and comprised 79 genes from six domestic animal breeds, running 0.85 seconds each time and with no false positives observed. Furthermore, it demonstrated excellent sensitivity and specificity when applied to this same dataset for another set of tests, with no true positives misidentified as false negatives.
Genomic CNV Wave was further evaluated against PennCNV, a standard CNV detection algorithm on the same dataset, using this comparison approach. Genatic cnv Wave proved more accurate at detecting deletions and duplications. Genomic CNV Wave was significantly superior in detecting duplications than other algorithms; for deletions it identified almost three times more events due to its ability to identify low-resolution duplications that may otherwise go undetected by other algorithms. Functional enrichment analysis was also conducted in order to assess biological significance; genes involved with CNVs included those related to olfactory receptors, G-protein coupled receptors and metabolic processes as enriched genes.
Detection of Heterozygous Deletions
Finding heterozygous deletions presents a daunting challenge for genome-wide sequencing as their breakpoints often lie within large genomic regions. Therefore, in order to detect CNVs with heterozygous alleles it requires an innovative method incorporating depth-of-coverage techniques that ensure accurate detection. This study created a novel CNV detection method using visualization of B-allele frequency (BAF) to detect small heterozygous deletions. This method takes advantage of the fact that satellite clusters containing duplication or deletion alleles appear as clearly differentiated clusters with increasing or decreasing signal intensities on scatterplots, thus making their characteristics useful in genotyping samples and thus identifying CNVs.
This method represents a major advancement over previous techniques that required manual expert review and experimental validation of original CNV calls, as its performance is comparable with an advanced neural network-based method for CNV detection from WGS data. Furthermore, this new technique is significantly more sensitive than standard genome-wide depth-of-coverage based CNV detection approaches.
To detect CNVs, this innovative method uses an iterative procedure utilizing genomic wave analysis, probe specific BAF, and population based haplotype information in an iterative fashion to filter out false positives and detect CNVs. Each genomic wave is generated by sequentially subdividing the target CNV region into smaller, equally sized 5′- and 3′-flanking regions. The genomic waves created are then visualized on scatterplots, where their median signal intensity is plotted against that of surrounding regions. This BAF distribution allows us to identify CNVs by their distinctive shape and size; their identification can then be confirmed through satellite clusters with high or low BAF intensities.
With PennCNV, we identified 1040 true CNVs comprising 267 duplication and 773 deletion B-alleles on six regions on chromosomes 2, 4, 5, 7, and 9 (Table 1). Figure 1 displays scatterplot-based calling of CNV on Chromosome 2. A suggested deletion not detected by PennCNV was confirmed through noise reduction of LRR values for relevant probes and visual inspection after scatterplot processing.
