Variant Calling Methods Research Articles

StructuralVariantAnnotation: a R/Bioconductor foundation for a caller-agnostic structural variant software ecosystem.

SummaryStructuralVariantAnnotation is an R/Bioconductor package that provides a framework for decoupling downstream analysis of structural variant breakpoints from upstream variant calling methods. It standardizes the representational format from BEDPE, or any of the three different notations supported by VCF into a breakpoint GRanges data structure suitable for use by the wider Bioconductor ecosystem. It handles both transitive breakpoints and duplication/insertion notational differences of identical variants—both common scenarios when comparing short/long read-based call sets that confound downstream analysis. StructuralVariantAnnotation provides the caller-agnostic foundation needed for a R/Bioconductor ecosystem of structural variant annotation, classification and interpretation tools able to handle both simple and complex genomic rearrangements.Availability and implementationStructuralVariantAnnotation is implemented in R and available for download as the Bioconductor StructuralVariantAnnotation package. Details can be found at https://www.bioconductor.org/packages/release/bioc/html/StructuralVariantAnnotation.html. It has been released under a GPL license.

Benchmarking small-variant genotyping in polyploids.

Genotyping from sequencing is the basis of emerging strategies in the molecular breeding of polyploid plants. However, compared with the situation for diploids, in which genotyping accuracies are confidently determined with comprehensive benchmarks, polyploids have been neglected; there are no benchmarks measuring genotyping error rates for small variants using real sequencing reads. We previously introduced a variant calling method, Octopus, that accurately calls germline variants in diploids and somatic mutations in tumors. Here, we evaluate Octopus and other popular tools on whole-genome tetraploid and hexaploid data sets created using in silico mixtures of diploid Genome in a Bottle (GIAB) samples. We find that genotyping errors are abundant for typical sequencing depths but that Octopus makes 25% fewer errors than other methods on average. We supplement our benchmarks with concordance analysis in real autotriploid banana data sets.

Evaluation of Germline Structural Variant Calling Methods for Nanopore Sequencing Data.

Structural variants (SVs) are genomic rearrangements that involve at least 50 nucleotides and are known to have a serious impact on human health. While prior short-read sequencing technologies have often proved inadequate for a comprehensive assessment of structural variation, more recent long reads from Oxford Nanopore Technologies have already been proven invaluable for the discovery of large SVs and hold the potential to facilitate the resolution of the full SV spectrum. With many long-read sequencing studies to follow, it is crucial to assess factors affecting current SV calling pipelines for nanopore sequencing data. In this brief research report, we evaluate and compare the performances of five long-read SV callers across four long-read aligners using both real and synthetic nanopore datasets. In particular, we focus on the effects of read alignment, sequencing coverage, and variant allele depth on the detection and genotyping of SVs of different types and size ranges and provide insights into precision and recall of SV callsets generated by integrating the various long-read aligners and SV callers. The computational pipeline we propose is publicly available at https://github.com/davidebolo1993/EViNCe and can be adjusted to further evaluate future nanopore sequencing datasets.

VC@Scale: Scalable and high-performance variant calling on cluster environments.

BackgroundRecently many new deep learning–based variant-calling methods like DeepVariant have emerged as more accurate compared with conventional variant-calling algorithms such as GATK HaplotypeCaller, Sterlka2, and Freebayes albeit at higher computational costs. Therefore, there is a need for more scalable and higher performance workflows of these deep learning methods. Almost all existing cluster-scaled variant-calling workflows that use Apache Spark/Hadoop as big data frameworks loosely integrate existing single-node pre-processing and variant-calling applications. Using Apache Spark just for distributing/scheduling data among loosely coupled applications or using I/O-based storage for storing the output of intermediate applications does not exploit the full benefit of Apache Spark in-memory processing. To achieve this, we propose a native Spark-based workflow that uses Python and Apache Arrow to enable efficient transfer of data between different workflow stages. This benefits from the ease of programmability of Python and the high efficiency of Arrow’s columnar in-memory data transformations.ResultsHere we present a scalable, parallel, and efficient implementation of next-generation sequencing data pre-processing and variant-calling workflows. Our design tightly integrates most pre-processing workflow stages, using Spark built-in functions to sort reads by coordinates and mark duplicates efficiently. Our approach outperforms state-of-the-art implementations by >2 times for the pre-processing stages, creating a scalable and high-performance solution for DeepVariant for both CPU-only and CPU + GPU clusters.ConclusionsWe show the feasibility and easy scalability of our approach to achieve high performance and efficient resource utilization for variant-calling analysis on high-performance computing clusters using the standardized Apache Arrow data representations. All codes, scripts, and configurations used to run our implementations are publicly available and open sourced; see https://github.com/abs-tudelft/variant-calling-at-scale.

Performance evaluation of pipelines for mapping, variant calling and interval padding, for the analysis of NGS germline panels

BackgroundNext-generation sequencing (NGS) represents a significant advancement in clinical genetics. However, its use creates several technical, data interpretation and management challenges. It is essential to follow a consistent data analysis pipeline to achieve the highest possible accuracy and avoid false variant calls. Herein, we aimed to compare the performance of twenty-eight combinations of NGS data analysis pipeline compartments, including short-read mapping (BWA-MEM, Bowtie2, Stampy), variant calling (GATK-HaplotypeCaller, GATK-UnifiedGenotyper, SAMtools) and interval padding (null, 50 bp, 100 bp) methods, along with a commercially available pipeline (BWA Enrichment, Illumina®). Fourteen germline DNA samples from breast cancer patients were sequenced using a targeted NGS panel approach and subjected to data analysis.ResultsWe highlight that interval padding is required for the accurate detection of intronic variants including spliceogenic pathogenic variants (PVs). In addition, using nearly default parameters, the BWA Enrichment algorithm, failed to detect these spliceogenic PVs and a missense PV in the TP53 gene. We also recommend the BWA-MEM algorithm for sequence alignment, whereas variant calling should be performed using a combination of variant calling algorithms; GATK-HaplotypeCaller and SAMtools for the accurate detection of insertions/deletions and GATK-UnifiedGenotyper for the efficient detection of single nucleotide variant calls.ConclusionsThese findings have important implications towards the identification of clinically actionable variants through panel testing in a clinical laboratory setting, when dedicated bioinformatics personnel might not always be available. The results also reveal the necessity of improving the existing tools and/or at the same time developing new pipelines to generate more reliable and more consistent data.

A unified haplotype-based method for accurate and comprehensive variant calling.

Almost all haplotype-based variant callers were designed specifically for detecting common germline variation in diploid populations, and give suboptimal results in other scenarios. Here we present Octopus, a variant caller that uses a polymorphic Bayesian genotyping model capable of modeling sequencing data from a range of experimental designs within a unified haplotype-aware framework. Octopus combines sequencing reads and prior information to phase-called genotypes of arbitrary ploidy, including those with somatic mutations. We show that Octopus accurately calls germline variants in individuals, including single nucleotide variants, indels and small complex replacements such as microinversions. Using a synthetic tumor data set derived from clean sequencing data from a sample with known germline haplotypes and observed mutations in a large cohort of tumor samples, we show that Octopus is more sensitive to low-frequency somatic variation, yet calls considerably fewer false positives than other methods. Octopus also outputs realigned evidence BAM files to aid validation and interpretation.

Evaluation of variant calling tools for large plant genome re-sequencing

BackgroundDiscovering single nucleotide polymorphisms (SNPs) from agriculture crop genome sequences has been a widely used strategy for developing genetic markers for several applications including marker-assisted breeding, population diversity studies for eco-geographical adaption, genotyping crop germplasm collections, and others. Accurately detecting SNPs from large polyploid crop genomes such as wheat is crucial and challenging. A few variant calling methods have been previously developed but they show a low concordance between their variant calls. A gold standard of variant sets generated from one human individual sample was established for variant calling tool evaluations, however hitherto no gold standard of crop variant set is available for wheat use. The intent of this study was to evaluate seven SNP variant calling tools (FreeBayes, GATK, Platypus, Samtools/mpileup, SNVer, VarScan, VarDict) with the two most popular mapping tools (BWA-mem and Bowtie2) on wheat whole exome capture (WEC) re-sequencing data from allohexaploid wheat.ResultsWe found the BWA-mem mapping tool had both a higher mapping rate and a higher accuracy rate than Bowtie2. With the same mapping quality (MQ) cutoff, BWA-mem detected more variant bases in mapping reads than Bowtie2. The reads preprocessed with quality trimming or duplicate removal did not significantly affect the final mapping performance in terms of mapped reads. Based on the concordance and receiver operating characteristic (ROC), the Samtools/mpileup variant calling tool with BWA-mem mapping of raw sequence reads outperformed other tests followed by FreeBayes and GATK in terms of specificity and sensitivity. VarDict and VarScan were the poorest performing variant calling tools with the wheat WEC sequence data.ConclusionThe BWA-mem and Samtools/mpileup pipeline, with no need to preprocess the raw read data before mapping onto the reference genome, was ascertained the optimum for SNP calling for the complex wheat genome re-sequencing. These results also provide useful guidelines for reliable variant identification from deep sequencing of other large polyploid crop genomes.

Abstract 1329: Identifying low allele frequency somatic variants using the Saphyr System

Abstract Cancer is known to be a dynamic disease with subclones evolving throughout the stages. These progresses create a sample of high heterogeneity, in which high throughput and high sensitivity structural variant calling sensitivity are essential to detect the low frequency events in order to have a thorough understanding of the cancer. Using the Bionano Saphyr® System, a genome imaging platform with a throughput of up to 1.3 Tbp per flowcell, sensitive detection of these low allele frequency structural rearrangements is facilitated by the Bionano data analysis tools, the Rare Variant Pipeline and Variant Annotation Pipeline. Here we compared three tumor-normal pairs of cell lines: the HCC2218, HCC1395 and HCC1954. We first collected about 1.3 Tbp for each of the six tumor and normal cell lines. Each dataset has molecule N50s above 270 kbp. These datasets were input into the Rare Variant Pipeline, where molecules deviating from the reference were locally assembled and structural variants were detected with the local assemblies. Running the Variant Annotation Pipeline then compared the structural variants between the tumor and the control samples and identified variants that are somatic in each pair. With the somatic calls found, we identified various numbers of low allele frequency events in each pair. For example, we observed a low frequency translocation t(2,12) that is unique to the HCC1954 tumor sample. This translocation involves the protein kinase ZAK, a gene known to take part in cell cycle checkpoint regulation (Tosti et al., 2004). In the HCC2218 tumor sample, we found a 381 kbp deletion at AUTS2 (autism susceptibility candidate 2), a gene which has also been associated with acute lymphoblastic leukemia (Denk et al., 2012) and other cancers (Stadler et al., 2012). These events are of low allele frequency and are found to be somatic. In conclusion, cancer samples are heterogeneous and thus, high throughput and sensitive structural variant calling methods are required for a complete examination of the structural mutations that may be present in tumor samples. Bionano has provided an efficient and streamlined data collection platform as well as a smooth and intuitive bioinformatics workflow for expeditious detection of low allele frequency events such as those herein illustrated. References Denk, Dagmar, et al. "PAX5-AUTS2: a recurrent fusion gene in childhood B-cell precursor acute lymphoblastic leukemia." Leukemia research 36.8 (2012): e178-e181. Stadler, Zsofia K., et al. "Rare de novo germline copy-number variation in testicular cancer." The American Journal of Human Genetics 91.2 (2012): 379-383. Tosti, Elena, et al. "The stress kinase MRK contributes to regulation of DNA damage checkpoints through a p38γ-independent pathway." Journal of Biological Chemistry 279.46 (2004): 47652-47660. Citation Format: Joyce Lee, Andy Wing Chun PANG, Caspar Groß, Jakob Admard, Elena Buena-Atienza, Stephan Ossowski, Thomas Anantharaman, Mark Oldakowski, Sven Bocklandt, Alex Hastie. Identifying low allele frequency somatic variants using the Saphyr System [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 1329.

A robust benchmark for detection of germline large deletions and insertions.

New technologies and analysis methods are enabling genomic structural variants (SVs) to be detected with ever-increasing accuracy, resolution, and comprehensiveness. To help translate these methods to routine research and clinical practice, we developed the first sequence-resolved benchmark set for identification of both false negative and false positive germline large insertions and deletions. To create this benchmark for a broadly consented son in a Personal Genome Project trio with broadly available cells and DNA, the Genome in a Bottle (GIAB) Consortium integrated 19 sequence-resolved variant calling methods from diverse technologies. The final benchmark set contains 12745 isolated, sequence-resolved insertion (7281) and deletion (5464) calls ≥50 base pairs (bp). The Tier 1 benchmark regions, for which any extra calls are putative false positives, cover 2.51 Gbp and 5262 insertions and 4095 deletions supported by ≥1 diploid assembly. We demonstrate the benchmark set reliably identifies false negatives and false positives in high-quality SV callsets from short-, linked-, and long-read sequencing and optical mapping.

Exploring the limit of using a deep neural network on pileup data for germline variant calling

Single-molecule sequencing technologies have emerged in recent years and revolutionized structural variant calling, complex genome assembly and epigenetic mark detection. However, the lack of a highly accurate small variant caller has limited these technologies from being more widely used. Here, we present Clair, the successor to Clairvoyante, a program for fast and accurate germline small variant calling, using single-molecule sequencing data. For Oxford Nanopore Technology data, Clair achieves better precision, recall and speed than several competing programs, including Clairvoyante, Longshot and Medaka. Through studying the missed variants and benchmarking intentionally overfitted models, we found that Clair may be approaching the limit of possible accuracy for germline small variant calling using pileup data and deep neural networks. Clair requires only a conventional central processing unit (CPU) for variant calling and is an open-source project available at https://github.com/HKU-BAL/Clair. A lack of accurate and efficient variant calling methods has held back single-molecule sequencing technologies from clinical applications. The authors present a deep-learning method for fast and accurate germline small variant calling, using single-molecule sequencing data.

Critical evaluation of copy number variant calling methods using DNA methylation.

Recent technological and methodological developments have enabled the use of array-based DNA methylation data to call copy number variants (CNVs). ChAMP, Conumee, and cnAnalysis450k are popular methods currently used to call CNVs using methylation data. However, so far, no studies have analyzed the reliability of these methods using real samples. Data from a cohort of individuals with genotype and DNA methylation data generated using the HumanMethylation450 and MethylationEPIC BeadChips were used to assess the consistency between the CNV calls generated by methylation and genotype data. We also took advantage of repeated measures of methylation data collected from the same individuals to compare the reliability of CNVs called by ChAMP, Conumee, and cnAnalysis450k for both the methylation arrays. ChAMP identified more CNVs than Conumee and cnAnalysis450k for both the arrays and, as a consequence, had a higher overlap (~62%) with the calls from the genotype data. However, all methods had relatively low reliability. For the MethylationEPIC array, Conumee had the highest reliability (57.6%), whereas for the HumanMethylation450 array, cnAnalysis450k had the highest reliability (43.0%). Overall, the MethylationEPIC array provided significant gains in reliability for CNV calling over the HumanMethylation450 array but not for overlap with CNVs called using genotype data.

Deep DNA Sequencing Identifies Small Mutations in Patients with Acute Myelogenous Leukemia (AML) and High-Risk Myelodysplastic Syndromes (MDS): Analysis of Intratumor Genomic Heterogeneity for Prediction of Disease Relapse

Background: Patients with unfavorable AML have an estimated 3-year overall survival (OS) of < 15% with standard chemotherapy treatment. The recommendation for consolidative allogeneic transplant can be challenging and is usually an individualized assessment. Accurate and sensitive genomic data may improve the decision making process. Detailed mutational analyses using myeloid-specific gene assays have shown to refine disease prognostication from standard six-gene mutational analysis (Gerstung et al. Nature Genetics2017, Dohner et al. Blood2017). Several studies using mutation analysis for assessing minimal residual disease (MRD) have been performed. Specifically, in two large studies, isolated presence of DNMT3A, TET2, or ASXL1mutations (associated with clonal hematopoiesis) was not prognostic but detection of other mutations, independent of pathogenicity, at >2.5% predicted relapse, relapse-free survival, and overall survival (Jongen-Lavrencic et al. JEJM2018, Morita et al. JCO2018). Of note, flow cytometry was also used to detect minimal residual disease (MRD), and at a sensitivity of 0.1% provided independent prognostic value; however even when used in combination, nearly 25% of relapses were not predicted. At Rutgers, we utilize deep DNA sequencing and real-time multiplex PCR assays to detect mutant alleles in the presence of 10,000 wild-type fragments. Here, we hypothesize that detailed analysis of available mutational data and detection of mutations at allele frequencies <2.5% in pre-treatment and sequentially collected samples during treatment serves as a targeted and ultrasensitive measure of clonal heterogeneity and may correspond better to prognosis and ultimately guide further treatment decisions based on MRD sequencing. Methods: We recruited 99 patients with newly diagnosed high-risk MDS and/or AML.Each patient had standard myeloid mutation profiles obtained at diagnosis (pre-treatment) and on the 'day 30' recovery bone marrow. Patient characteristics and bone marrow biopsy results were obtained including standard cytogenetics and conventional molecular studies. Additional and contemporaneous analyses were performed for 18 patients, diagnosed and treated from February 2017 through July 2019,utilizinga 49-gene panel (RainDance Technologies) on Illumina MiSeq at high depth (2,000x-20,000x). We identified all sites different from the reference using an inclusive variant caller and used a Bayesian approach to detect true mutations against mutation-specific background error at >0.1% allele frequencies. We tested the variants against previously sequenced non-malignant samples and guided the analysis by detected mutations in longitudinal data (Hadigol et al. BMC Bioinformatics2018, Rabadan et al. J Stat Phys2018). Results:The median age of patients with deep-sequencing data was 55 years (range, 26-73 [n=4 male; n=14 female]). There were 14 patients with AML and 4 patients with high-risk MDS. Sixty-one percent of patients received standard 7+3 induction therapy and 14 patients received consolidative allogeneic stem cell transplant (SCT), 93% of whom received it at first complete remission (CR1). With the median follow-up of 19 months (range, 12-27), the median time to relapse in 5 patients (28%) was 13 months (range, 8-15). In the relapsed patients, there was a median of 1 mutation (range, 0-2) reported by conventional variant calling methods at CR1. However, Bayesian deep-sequencing analysis identified additional non-synonymous coding mutations (range, 1-4) at allele frequencies <2.5% in 4 of these patients for whom sequencing data was available at this time point (Table). It is important to highlight that some of these mutations were not detected at initial diagnostic sample and could be present in clones with higher fitness after treatment. Conclusion: Our approach identified new and relevant mutations that were not reported with conventional genomic analysis at a crucial time point in the measurement of treatment response in AML and high-risk MDS. Data such as these may refine prognostication and ultimately be used as to help guide further treatment decision making for patients. Continued and larger prospective analyses with novel deep-sequencing technology are needed to better inform outcomes for AML and high-risk MDS patients. Disclosures No relevant conflicts of interest to declare.

Longshot enables accurate variant calling in diploid genomes from single-molecule long read sequencing

Whole-genome sequencing using sequencing technologies such as Illumina enables the accurate detection of small-scale variants but provides limited information about haplotypes and variants in repetitive regions of the human genome. Single-molecule sequencing (SMS) technologies such as Pacific Biosciences and Oxford Nanopore generate long reads that can potentially address the limitations of short-read sequencing. However, the high error rate of SMS reads makes it challenging to detect small-scale variants in diploid genomes. We introduce a variant calling method, Longshot, which leverages the haplotype information present in SMS reads to accurately detect and phase single-nucleotide variants (SNVs) in diploid genomes. We demonstrate that Longshot achieves very high accuracy for SNV detection using whole-genome Pacific Biosciences data, outperforms existing variant calling methods, and enables variant detection in duplicated regions of the genome that cannot be mapped using short reads.

Targeted variant detection using unaligned RNA-Seq reads.

Mutations identified in acute myeloid leukemia patients are useful for prognosis and for selecting targeted therapies. Detection of such mutations using next-generation sequencing data requires a computationally intensive read mapping step followed by several variant calling methods. Targeted mutation identification drastically shifts the usual tradeoff between accuracy and performance by concentrating all computations over a small portion of sequence space. Here, we present km, an efficient approach leveraging k-mer decomposition of reads to identify targeted mutations. Our approach is versatile, as it can detect single-base mutations, several types of insertions and deletions, as well as fusions. We used two independent cohorts (The Cancer Genome Atlas and Leucegene) to show that mutation detection by km is fast, accurate, and mainly limited by sequencing depth. Therefore, km allows the establishment of fast diagnostics from next-generation sequencing data and could be suitable for clinical applications.

28. KARYOMAPPING USING WHOLE GENOME SEQUENCING DATA: PRECLINICAL VALIDATION OF A PERSONALISED APPROACH TO PGT-M

Introduction Karyomapping utilises microarray technology to track informative single nucleotide polymorphisms (SNPs) and assign haplotypes to embryos without the need for direct mutation detection (Handyside et al. 2010). Although suitable for most couples undergoing PGT-M, the use of a fixed set of SNPs may provide inadequate information in certain genetic regions or in consanguineous couples (Konstantinidis et al. 2015). Whole genome sequencing (WGS) has the potential to individualise informative SNP detection and perform concurrent direct mutation detection. However, no studies have evaluated the potential for application of the karyomapping algorithm to WGS data. Material and Methods The Platinum Genomes dataset ( https://sapac.illumina.com/platinumgenomes.html ) was analysed as a model for the application of karyomapping to WGS data. Variant call format (VCF) files were analysed according to the principles of karyomapping to create a map of informative biallelic SNPs in each grandparental haplotype. A model of embryo biopsy was created by diluting DNA from one of the family members (NA12878) to 30pg and whole genome amplification was performed with the Repli-G single cell kit (Qiagen, Hilden, Germany) as a model of trophectoderm biopsy. Whole genome sequencing was performed in triplicate on a high output NextSeq 2500 run (Illumina, California, USA) at average 15x coverage and the number of informative SNPs detected compared to the karyomapping SNP microarray. Comparison of variant calling methods was performed using the Genome Analysis Toolkit v3.4 (Broad Institute, Massachusetts, USA) and DeepVariant (Google, California, USA). Results Analysis of WGS data from the Platinum Genomes family revealed a total of 2,338,158 informative SNPs distributed evenly between grandparental haplotypes, of which 85,053 (3.6%) would have been detectable using the karyomapping SNP array. Analysis of the karyomapping window around common genes requiring PGT-M demonstrated an average 32x increase in the number of targets available for analysis in a PGT-M cycle. Analysis of 15x coverage WGS data in a model of embryo biopsy demonstrated an average 279,851 informative SNPs detected using standard variant calling and an average 285,107 using AI-based variant calling. The median distance between informative SNPs was reduced from 26,782 nucleotides to 1,212 nucleotides when comparing WGS-based karyomapping to SNP array-based karyomapping. Conclusions This study provides preclinical validation for the application of the karyomapping algorithm to WGS data. The markedly increased number of informative SNPs will allow successful feasibility studies for all couples presenting for PGT-M. The decreased distance between informative SNPs detected in a model embryo biopsy will improve diagnostic accuracy, particularly in detecting small double crossover events and reducing the number of cycles in which inadequate informative markers have been detected in the embryo sample. Finally, the use of WGS for diagnostic purposes, especially for rare and undiagnosed genetic disease, will allow the creation of a high resolution digital karyomap for couples with an affected child and improve the transition to IVF and PGT-M.

Abstract 1682: Non-V600 BRAF mutants in melanoma: Multiple MAPK mutations and distinctive RNA expression

Abstract Large discovery sequencing analyses conducted by our group and others enabled classification of melanomas into major and minor subtypes, helping to elucidate the role and diversity of activating mutations within the MAPK pathway. We sequenced 482 melanoma PDX, cell lines and patient biopsies, of which 20 (4.1%) carry non-V600 BRAF mutations. Of the 20 mutations, 14 are pathogenic, 5 are likely pathogenic and 1 is a variant of unknown significance. Interestingly, 78.5% of tumors with pathogenic non-V600 mutants and all likely pathogenic or VUS non-V600 mutants carry additional mutations within the MAPK pathway. To explore this pattern, we queried publicly available melanoma sequencing data from six previously published sequencing studies of melanoma. Consistent with our findings, nonV600-BRAF mutations were identified in 0 - 9% of melanoma sequencing studies. Of the tumors carrying non-V600 BRAF mutants, 26 of 42 (62%) carry additional mutations in the MAPK pathway. Given the co-occurrence of MAPK mutations with non-V600 mutants, we considered whether these mutations are found together within the same cell or if mutants are encoded in distinct cell populations and reflect admixture of a heterogenous tumor. We compared the variant allele frequency (VAF) of co-occurring MAPK mutants with their non-V600 BRAF counterparts in 41 available samples. The VAF between mutants was discordant (>0.10 change in VAF between mutants) in 27 of 41 (68%) of the samples. To more precisely assess variants at single-cell resolution, we established a PCR based method for variant calling using single-cell-derived cDNA libraries. Pilot studies supported distinct expression patterns and we are working to generate robust and reproducible sequencing data. We also compared RNA expression data between BRAFV600 melanomas and non-V600 melanomas. This analysis revealed distinct transcriptomes that were easily distinguished based on previously established quaternary structure and kinase activity of various BRAF mutants. Interestingly, MAPK-pathway genes did not account for most of the differential expression, and an AXL-high/MITF-low expression program was highly enriched in non-V600 BRAF tumors relative to their V600 BRAF counterpart (FDR <0.01). Taken together, non-V600 BRAF mutant melanomas represent a subtype characterized by a tendency for multiple MAPK mutations and an AXL-high/MITF-low enriched RNA expression pattern differentiated based on BRAF structure and function. Further studies of this rare sub-type of melanoma are important to demonstrate the diversity and interplay of MAPK dysregulation in cancer and may reveal mechanisms of tumorigenesis. Citation Format: Elizabeth Kiernan, Paul Bastian, Liza Dorfman, Ioannis N. Anastopoulos, Brad Wubbenhorst, Clemens Krepler, Vito Rebecca, Meenhard Herlyn, Katherine Nathanson. Non-V600 BRAF mutants in melanoma: Multiple MAPK mutations and distinctive RNA expression [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1682.

Best practices for benchmarking germline small-variant calls in human genomes.

Standardized benchmarking approaches are required to assess the accuracy of variants called from sequence data. Although variant-calling tools and the metrics used to assess their performance continue to improve, important challenges remain. Here, as part of the Global Alliance for Genomics and Health (GA4GH), we present a benchmarking framework for variant calling. We provide guidance on how to match variant calls with different representations, define standard performance metrics, and stratify performance by variant type and genome context. We describe limitations of high-confidence calls and regions that can be used as truth sets (for example, single-nucleotide variant concordance of two methods is 99.7% inside versus 76.5% outside high-confidence regions). Our web-based app enables comparison of variant calls against truth sets to obtain a standardized performance report. Our approach has been piloted in the PrecisionFDA variant-calling challenges to identify the best-in-class variant-calling methods within high-confidence regions. Finally, we recommend a set of best practices for using our tools and evaluating the results.

Vi-HMM: a novel HMM-based method for sequence variant identification in short-read data

BackgroundAccurate and reliable identification of sequence variants, including single nucleotide polymorphisms (SNPs) and insertion-deletion polymorphisms (INDELs), plays a fundamental role in next-generation sequencing (NGS) applications. Existing methods for calling these variants often make simplified assumptions of positional independence and fail to leverage the dependence between genotypes at nearby loci that is caused by linkage disequilibrium (LD).Results and conclusionWe propose vi-HMM, a hidden Markov model (HMM)-based method for calling SNPs and INDELs in mapped short-read data. This method allows transitions between hidden states (defined as “SNP,” “Ins,” “Del,” and “Match”) of adjacent genomic bases and determines an optimal hidden state path by using the Viterbi algorithm. The inferred hidden state path provides a direct solution to the identification of SNPs and INDELs. Simulation studies show that, under various sequencing depths, vi-HMM outperforms commonly used variant calling methods in terms of sensitivity and F1 score. When applied to the real data, vi-HMM demonstrates higher accuracy in calling SNPs and INDELs.

SmCounter2: an accurate low-frequency variant caller for targeted sequencing data with unique molecular identifiers.

MotivationLow-frequency DNA mutations are often confounded with technical artifacts from sample preparation and sequencing. With unique molecular identifiers (UMIs), most of the sequencing errors can be corrected. However, errors before UMI tagging, such as DNA polymerase errors during end repair and the first PCR cycle, cannot be corrected with single-strand UMIs and impose fundamental limits to UMI-based variant calling.ResultsWe developed smCounter2, a UMI-based variant caller for targeted sequencing data and an upgrade from the current version of smCounter. Compared to smCounter, smCounter2 features lower detection limit that decreases from 1 to 0.5%, better overall accuracy (particularly in non-coding regions), a consistent threshold that can be applied to both deep and shallow sequencing runs, and easier use via a Docker image and code for read pre-processing. We benchmarked smCounter2 against several state-of-the-art UMI-based variant calling methods using multiple datasets and demonstrated smCounter2’s superior performance in detecting somatic variants. At the core of smCounter2 is a statistical test to determine whether the allele frequency of the putative variant is significantly above the background error rate, which was carefully modeled using an independent dataset. The improved accuracy in non-coding regions was mainly achieved using novel repetitive region filters that were specifically designed for UMI data.Availability and implementationThe entire pipeline is available at https://github.com/qiaseq/qiaseq-dna under MIT license.Supplementary information Supplementary data are available at Bioinformatics online.

Abstract 926: Multi-Center Mutation Calling in Multiple Cancers: The MC3 Project

Abstract The Multi-Center Mutation Calling in Multiple Cancers (MC3) data set provides consistent variant calling and filtering across the 10K patients in The Cancer Genome Atlas (TCGA). The MC3 was a collaborative science effort, driven by a consortium of researchers across multiple institutions, to form a TCGA capstone project focused on cross-tumor type automated analysis. This dataset covers 33 different cancer types using an ensemble of 7 advanced mutation-calling algorithms with scoring and artifact filtering, implemented for sharing in a reproducible, portable, standardized workflow. The resulting dataset represents several million core-hours of computational time on over 400 TB of short read data using the current state-of-the-art variant calling and filtering methods. In the past decade, the precipitous drop in sequencing cost from $10M to $1000 has allowed for larger cohorts of data to be analyzed. In 2016 there were an estimated 1.6M new cancer diagnoses in the United States. Scaling computational systems and genomic analysis to work at this scale requires the coordination of many institutions, many experiments and many computational techniques. Aside from problems of scale, there are several issues that prevent large analyses: 1) deployment of reproducible computing methods in new computing environments, 2) the ability to deploy methods without manual intervention, 3) the biases of single methods and the need for consensus, and 4) the large amount of noise and false positives that come from data including both germline sequencing, heterogeneous tumor sequencing, technical artifacts, and low variant allele fraction of observed reads. The MC3 project dealt with these issues by 1) modeling the pipeline using Common Workflow Language (CWL) format with the required software packages deployed using Docker software containerization technology, ensuring the ability to deploy analysis on new computer systems; 2) crafting the entire pipeline using best practice parameters and applying them consistently, across the entire cohort; we have also analyzed validation data to determine particular cancer type where these parameters do not fit; 3) deploying an ensemble of 7 variant calling methods including MuTect, MuSE, Radia, Somatic Sniper, Pindel, Indelocator and Varscan2 (both indel and SNP calling); and 4) applying a robust set of additional filters, as well as applying meta-calling based on results of multiple callers, to reduce false-positive rates. Over 20 million variants were produced and the data generated by this work have formed the basis of the somatic exome variant analysis presented in the other papers from the TCGA PanCanAtlas project. A set of over 3 million high-quality variants from a public release of this data has been made available so that researchers may easily use this open resource. Citation Format: Kyle Ellrott, Mathew Bailey, Gordon Saksena, Kyle Covington, Cyriac Kandoth, Chip Stewart, Michael McLellan, Heidi Sofia, Carolyn Hutter, Gad Getz, David Wheeler, Li Ding. Multi-Center Mutation Calling in Multiple Cancers: The MC3 Project [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 926.