Search

The lack of samples for generating standardized DNA datasets for setting up a sequencing pipeline or benchmarking the performance of different algorithms limits the implementation and uptake of cancer genomics. Here, we describe reference call sets obtained from paired tumor–normal genomic DNA (gDNA) samples derived from a breast cancer cell line—which is highly heterogeneous, with an aneuploid genome, and enriched in somatic alterations—and a matched lymphoblastoid cell line. We partially validated both somatic mutations and germline variants in these call sets via whole-exome sequencing (WES) with different sequencing platforms and targeted sequencing with >2,000-fold coverage, spanning 82% of genomic regions with high confidence. Although the gDNA reference samples are not representative of primary cancer cells from a clinical sample, when setting up a sequencing pipeline, they not only minimize potential biases from technologies, assays and informatics but also provide a unique resource for benchmarking ‘tumor-only’ or ‘matched tumor–normal’ analyses.

The precisionFDA Truth Challenge V2 aimed to assess the state of the art of variant calling in challenging genomic regions. Starting with FASTQs, 20 challenge participants applied their variant-calling pipelines and submitted 64 variant call sets for one or more sequencing technologies (Illumina, PacBio HiFi, and Oxford Nanopore Technologies). Submissions were evaluated following best practices for benchmarking small variants with updated Genome in a Bottle benchmark sets and genome stratifications. Challenge submissions included numerous innovative methods, with graph-based and machine learning methods scoring best for short-read and long-read datasets, respectively. With machine learning approaches, combining multiple sequencing technologies performed particularly well. Recent developments in sequencing and variant calling have enabled benchmarking variants in challenging genomic regions, paving the way for the identification of previously unknown clinically relevant variants.

BackgroundReproducible detection of inherited variants with whole genome sequencing (WGS) is vital for the implementation of precision medicine and is a complicated process in which each step affects variant call quality. Systematically assessing reproducibility of inherited variants with WGS and impact of each step in the process is needed for understanding and improving quality of inherited variants from WGS.ResultsTo dissect the impact of factors involved in detection of inherited variants with WGS, we sequence triplicates of eight DNA samples representing two populations on three short-read sequencing platforms using three library kits in six labs and call variants with 56 combinations of aligners and callers. We find that bioinformatics pipelines (callers and aligners) have a larger impact on variant reproducibility than WGS platform or library preparation. Single-nucleotide variants (SNVs), particularly outside difficult-to-map regions, are more reproducible than small insertions and deletions (indels), which are least reproducible when > 5 bp. Increasing sequencing coverage improves indel reproducibility but has limited impact on SNVs above 30×.ConclusionsOur findings highlight sources of variability in variant detection and the need for improvement of bioinformatics pipelines in the era of precision medicine with WGS.

Investigation of large structural variants (SVs) is a challenging yet important task in understanding trait differences in highly repetitive genomes. Combining different bioinformatic approaches for SV detection, we analyzed whole-genome sequencing data from 3000 rice genomes and identified 63 million individual SV calls that grouped into 1.5 million allelic variants. We found enrichment of long SVs in promoters and an excess of shorter variants in 5′ UTRs. Across the rice genomes, we identified regions of high SV frequency enriched in stress response genes. We demonstrated how SVs may help in finding causative variants in genome-wide association analysis. These new insights into rice genome biology are valuable for understanding the effects SVs have on gene function, with the prospect of identifying novel agronomically important alleles that can be utilized to improve cultivated rice.

Background Oncopanel genomic testing, which identifies important somatic variants, is increasingly common in medical practice and especially in clinical trials. Currently, there is a paucity of reliable genomic reference samples having a suitably large number of pre-identified variants for properly assessing oncopanel assay analytical quality and performance. The FDA-led Sequencing and Quality Control Phase 2 (SEQC2) consortium analyze ten diverse cancer cell lines individually and their pool, termed Sample A, to develop a reference sample with suitably large numbers of coding positions with known (variant) positives and negatives for properly evaluating oncopanel analytical performance. Results In reference Sample A, we identify more than 40,000 variants down to 1% allele frequency with more than 25,000 variants having less than 20% allele frequency with 1653 variants in COSMIC-related genes. This is 5–100× more than existing commercially available samples. We also identify an unprecedented number of negative positions in coding regions, allowing statistical rigor in assessing limit-of-detection, sensitivity, and precision. Over 300 loci are randomly selected and independently verified via droplet digital PCR with 100% concordance. Agilent normal reference Sample B can be admixed with Sample A to create new samples with a similar number of known variants at much lower allele frequency than what exists in Sample A natively, including known variants having allele frequency of 0.02%, a range suitable for assessing liquid biopsy panels. Conclusion These new reference samples and their admixtures provide superior capability for performing oncopanel quality control, analytical accuracy, and validation for small to large oncopanels and liquid biopsy assays.

SummaryThe precisionFDA Truth Challenge V2 aimed to assess the state-of-the-art of variant calling in difficult-to-map regions and the Major Histocompatibility Complex (MHC). Starting with FASTQ files, 20 challenge participants applied their variant calling pipelines and submitted 64 variant callsets for one or more sequencing technologies (~35X Illumina, ~35X PacBio HiFi, and ~50X Oxford Nanopore Technologies). Submissions were evaluated following best practices for benchmarking small variants with the new GIAB benchmark sets and genome stratifications. Challenge submissions included a number of innovative methods for all three technologies, with graph-based and machine-learning methods scoring best for short-read and long-read datasets, respectively. New methods out-performed the 2016 Truth Challenge winners, and new machine-learning approaches combining multiple sequencing technologies performed particularly well. Recent developments in sequencing and variant calling have enabled benchmarking variants in challenging genomic regions, paving the way for the identification of previously unknown clinically relevant variants.

Tumor Mutational Burden (TMB) is a measure of the abundance of somatic mutations in a tumor, which has been shown to be an emerging biomarker for both anti-PD-(L)1 treatment and prognosis. Nevertheless, multiple challenges still hinder the adoption of TMB for clinical decision-making. The key challenges are the inconsistency of TMB measurement among assays and a lack of meaningful threshold for TMB classification. We describe a powerful and flexible statistical framework for estimation and classification of TMB (ecTMB). ecTMB uses an explicit background mutation model to predict TMB robustly and consistently. In addition to the two known TMB subtypes, TMB-high and TMB-low, we discovered a novel TMB subtype, named TMB-extreme, which was significantly associated with patient survival outcome. This discovery enabled ecTMB to classify samples to biologically and clinically relevant subtypes defined by TMB.

RNA-sequencing (RNA-seq) is an essential technique for transcriptome studies, hundreds of analysis tools have been developed since it was debuted. Although recent efforts have attempted to assess the latest available tools, they have not evaluated the analysis workflows comprehensively to unleash the power within RNA-seq. Here we conduct an extensive study analysing a broad spectrum of RNA-seq workflows. Surpassing the expression analysis scope, our work also includes assessment of RNA variant-calling, RNA editing and RNA fusion detection techniques. Specifically, we examine both short- and long-read RNA-seq technologies, 39 analysis tools resulting in ~120 combinations, and ~490 analyses involving 15 samples with a variety of germline, cancer and stem cell data sets. We report the performance and propose a comprehensive RNA-seq analysis protocol, named RNACocktail, along with a computational pipeline achieving high accuracy. Validation on different samples reveals that our proposed protocol could help researchers extract more biologically relevant predictions by broad analysis of the transcriptome., RNA-seq is widely used for transcriptome analysis. Here, the authors analyse a wide spectrum of RNA-seq workflows and present a comprehensive analysis protocol named RNACocktail as well as a computational pipeline leveraging the widely used tools for accurate RNA-seq analysis.

Accurate detection of somatic mutations is challenging but critical to the understanding of cancer formation, progression, and treatment. We recently proposed NeuSomatic, the first deep convolutional neural network based somatic mutation detection approach and demonstrated performance advantages on in silico data. In this study, we used the first comprehensive and well-characterized somatic reference samples from the SEQC-II consortium to investigate best practices for utilizing deep learning framework in cancer mutation detection. Using the high-confidence somatic mutations established for these reference samples by the consortium, we identified strategies for building robust models on multiple datasets derived from samples representing real scenarios. The proposed strategies achieved high robustness across multiple sequencing technologies such as WGS, WES, AmpliSeq target sequencing for fresh and FFPE DNA input, varying tumor/normal purities, and different coverages (ranging from 10× - 2000×). NeuSomatic significantly outperformed conventional detection approaches in general, as well as in challenging situations such as low coverage, low mutation frequency, DNA damage, and difficult genomic regions.

We characterized two reference samples for NGS technologies: a human triple-negative breast cancer cell line and a matched normal cell line. Leveraging several whole-genome sequencing (WGS) platforms, multiple sequencing replicates, and orthogonal mutation detection bioinformatics pipelines, we minimized the potential biases from sequencing technologies, assays, and informatics. Thus, our “truth sets” were defined using evidence from 21 repeats of WGS runs with coverages ranging from 50X to 100X (a total of 140 billion reads). These “truth sets” present many relevant variants/mutations including 193 COSMIC mutations and 9,016 germline variants from the ClinVar database, nonsense mutations inBRCA1/2and missense mutations inTP53andFGFR1.Independent validation in three orthogonal experiments demonstrated a successful stress test of the truth set. We expect these reference materials and “truth sets” to facilitate assay development, qualification, validation, and proficiency testing. In addition, our methods can be extended to establish new fully characterized reference samples for the community.

Accurate detection of somatic mutations is still a challenge in cancer analysis. Here we present NeuSomatic, the first convolutional neural network approach for somatic mutation detection, which significantly outperforms previous methods on different sequencing platforms, sequencing strategies, and tumor purities. NeuSomatic summarizes sequence alignments into small matrices and incorporates more than a hundred features to capture mutation signals effectively. It can be used universally as a stand-alone somatic mutation detection method or with an ensemble of existing methods to achieve the highest accuracy., Somatic mutations are crucial to the understanding of cancer genesis, progression, and treatment, but are still challenging to detect. Here the authors present NeuSomatic, a convolutional neural network approach for accurate somatic mutation detection across various sequencing scenarios.

e14074 Background: Accurate and comprehensive interpretation of genomic variants has become a bottleneck in clinical sequencing applications due to the accelerated implementation of precision oncology and the rapid growth of relevant biomedical findings. We therefore are motivated to build Ephesus, a framework enabling curation of clinical evidence for biomarkers in somatic cancers. Currently Ephesus is the primary content source for Roche NAVIFY Mutation Profiler (NMP). Methods: Ephesus’ data model ensures adherence to best practice in clinical genomic content curation. Variant classification followed the AMP guidelines for somatic variant interpretation. Approvals and recommendations from regulatory agencies and organizations (FDA, NCCN, etc.) are applied on a regional basis. The data model is mapped to a set of PostgreSQL relational tables. Strategies such as data normalization according to standard ontologies, a submit-review-approval workflow with change logs, and biologically relevant datatype constraints are adopted to enforce data integrity. Ephesus is deployed as a web application, of which the UI allows users to conduct edits, filtering, sorting and bulk operations on entities by attributes. To maximize curation efficiency, a number of inference rules are applied before the content is ingested into NMP. Results: Currently Ephesus is developed to guide data collection, browsing and summarizing related to these primary entities: Biomarkers, Evidence items, Genes, Variants, Variant groups, and Drugs. The content exported on 2020/01/27 contains > 170K biomarker profiles (including combinations), in the context of 13 major cancer types. To evaluate the clinical reporting value of the curated knowledge, more than 60k real clinical cancer patient samples from the AACR GENIE project were queried against Ephesus. We compared the results to the latest content from several major knowledgebases. We see that the performance of Ephesus/NMP excels the others. Conclusions: We have developed Ephesus, a practical content curation framework which provides a highly structured and user-friendly environment for clinically relevant somatic biomarkers. The framework improved the productivity and quality of the manual curation, met clinical interpretation scope and complexity, and assures data integrity and auditability. [Table: see text]

The human genome is generally organized into stable chromosomes, and only tumor cells are known to accumulate kilobase (kb)-sized extrachromosomal circular DNA elements (eccDNAs). However, it must be expected that kb eccDNAs exist in normal cells as a result of mutations. Here, we purify and sequence eccDNAs from muscle and blood samples from 16 healthy men, detecting ~100,000 unique eccDNA types from 16 million nuclei. Half of these structures carry genes or gene fragments and the majority are smaller than 25 kb. Transcription from eccDNAs suggests that eccDNAs reside in nuclei and recurrence of certain eccDNAs in several individuals implies DNA circularization hotspots. Gene-rich chromosomes contribute to more eccDNAs per megabase and the most transcribed protein-coding gene in muscle, TTN (titin), provides the most eccDNAs per gene. Thus, somatic genomes are rich in chromosome-derived eccDNAs that may influence phenotypes through altered gene copy numbers and transcription of full-length or truncated genes., Somatic cells can accumulate structural variations such as deletions. Here, Møller et al. show that normal human cells generate large extrachromosomal circular DNAs (eccDNAs), most likely the products of excised DNA, that can be transcriptionally active and, thus, may have phenotypic consequences.

We present NeuSomatic, the first convolutional neural network approach for somatic mutation detection, which significantly outperforms previous methods on different sequencing platforms, sequencing strategies, and tumor purities. NeuSomatic summarizes sequence alignments into small matrices and incorporates more than a hundred features to capture mutation signals effectively. It can be used universally as a stand-alone somatic mutation detection method or with an ensemble of existing methods to achieve the highest accuracy.

Tumor Mutational Burden (TMB) is a measure of the abundance of somatic mutations in a tumor, which has been shown to be an emerging biomarker for anti-PD-(L)1 treatment. Nevertheless, multiple challenges still hinder the adoption of TMB for clinical decision-making. The current standard for TMB measurement requires counting the non-synonymous somatic mutations in a paired tumor-normal sample with whole-exome sequencing (WES); however, clinical diagnostics based on sequencing technologies still rely heavily on targeted panel sequencing. Thus, the key challenge is the inconsistency of the panel-based TMB measurement when compared to that of a more comprehensive WES. Targeted panel sequencing might overestimate TMB due to its enrichment of driver mutations and mutation hot spots. Another non-negligible challenge is the relatively arbitrary selection of the TMB-high cutoff. In order to improve the robustness of TMB measurement, we propose a novel TMB prediction method, ecTMB, based on a statistical model with a Bayesian framework. The model takes into account the heterogeneous mutation context in cancer, and other influencing factors, to estimate the sample- and gene-specific background mutation rates, which can systematically reduce driver-mutation impacts and include synonymous mutations in the estimation. The in-silico assessment of TMB estimation for the FoundationOne CDxTM panel by ecTMB showed an improved agreement between the WES-based TMB and the panel-based TMB with a slope close to 1 in Deming regression, whereas the standard counting method had a slope of 1.3. Further analysis of colorectal (n = 520), stomach (n = 435), and endometrial (n = 538) cancers on WES-based TMB using log transformation revealed three hidden cancer subtypes: TMB-low, TMB-high, and a novel subtype - TMB-extreme. Not only did these three cancer subtypes have distinct mutation profiles and tumor-infiltrating immune cell populations, but also they associated with patient survival outcomes significantly. We observed that TMB-high and TMB-extreme are associated with improved patient survival at different levels after considering age and cancer stage (hazard ratio (HR) for TMB-high = 0.8 with p-value = 0.1; hazard ratio (HR) for TMB-extreme = 0.32 with p-value = 0.006). We then extended ecTMB with a Gaussian Mixture Model to classify samples by the aforementioned cancer subtypes. Our results demonstrated that ecTMB could estimate panel-based TMB and classify samples based on TMB more robustly. Results also showed that the novel TMB-extreme subtype had a significantly lower hazard ratio than TMB-high, indicating better survival rate. We believe our method and also can help facilitate the adoption of TMB testing in clinical diagnostics. Citation Format: Lijing Yao, Marghoob Mohiyuddin, Hugo Lam. ecTMB: A robust method to estimate and classify tumor mutational burden [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 LB-213.

ROC curves for One-class classification using the L1 Distance, treating the 4000 random regions as negatives and the Spiral Genetics insertions calls as positives. (A) ROC curves for one-class models for each dataset separately and for all combined. (B) ROC curves for one-class model requiring 1 or more, 2 or more, 3 or more, or all 4 technologies to have high classification scores. See original data at https://plot.ly/~desuchen0929/386, and https://plot.ly/330/~parikhhm/. (PDF 351 kb)

Output format of svclassify. (DOC 30 kb)

Selected characteristics for hierarchical clustering and One-class models. (DOC 23 kb)

ROC curves for One-class classification using SVM and L1 â 3 or moreâ strategy, treating the 4000 random regions as training negatives, treating (A) the Personalis deletion calls and (B) the 1000 Genomes deletion calls as testing positives and treating the 2306 random regions as testing negatives. See original data at https://plot.ly/266/~parikhhm/, and https://plot.ly/274/~parikhhm/. (PDF 305 kb)

ROC curves for One-class classification using SVM, treating the 4000 Random regions as negatives and the Personalis or 1000 Genomes calls as positives. (A) ROC curves for one-class models for each dataset separately and for all combined for the Personalis validated deletion calls. (B) ROC curves for one-class models for each dataset separately and for all combined for the 1000 Genomes validated deletion calls. (C) ROC curves for one-class model requiring 1 or more, 2 or more, 3 or more, or all 4 technologies to have high classification scores for the Personalis validated deletion calls. (D) ROC curves for one-class model requiring 1 or more, 2 or more, 3 or more, or all 4 technologies to have high classification scores for the 1000 Genomes validated deletion calls. See original data at https://plot.ly/233/~parikhhm/, https://plot.ly/242/~parikhhm/, https://plot.ly/250/~parikhhm/, and https://plot.ly/258/~parikhhm/. (PDF 579 kb)

Histogram of Ď score for 2306 random regions with same size distribution as Personalis SVs. (PDF 322 kb)

ROC curves for One-class classification using the L1 Distance, treating the 2306 Random regions (size distribution matching to Personalis) as negatives and the Personalis as positives using the Platinum down-sampling 30x data. See original data at https://plot.ly/165/~parikhhm/. (PDF 169 kb)

ROC curves for One-class classification using SVM, treating the 4000 random regions as negatives and the Spiral Genetics insertions calls as positives. (A) ROC curves for one-class models for each dataset separately and for all combined. (B) ROC curves for one-class model requiring 1 or more, 2 or more, 3 or more, or all 4 technologies to have high classification scores. See original data at https://plot.ly/~desuchen0929/391, and https://plot.ly/297/~parikhhm/. (PDF 351 kb)

A many-core processor design for high-performance systems draws from embedded computing's low-power architectures and design processes, providing a radical alternative to cluster solutions.

A high-confidence, comprehensive human variant set is critical in assessing accuracy of sequencing algorithms, which are crucial in precision medicine based on high-throughput sequencing. Although recent works have attempted to provide such a resource, they still do not encompass all major types of variants including structural variants (SVs). Thus, we leveraged the massive high-quality Sanger sequences from the HuRef genome to construct by far the most comprehensive gold set of a single individual, which was cross validated with deep Illumina sequencing, population datasets and well-established algorithms. It was a necessary effort to completely reanalyze the HuRef genome as its previously published variants were mostly reported five years ago, suffering from compatibility, organization and accuracy issues that prevent their direct use in benchmarking. Our extensive analysis and validation resulted in a gold set with high specificity and sensitivity. In contrast to the current gold sets of the NA12878 or HS1011 genomes, our gold set is the first that includes small variants, deletion SVs and insertion SVs up to a hundred thousand base-pairs. We demonstrate the utility of our HuRef gold set to benchmark several published SV detection tools.

SomaticSeq is an accurate somatic mutation detection pipeline implementing a stochastic boosting algorithm to produce highly accurate somatic mutation calls for both single nucleotide variants and small insertions and deletions. The workflow currently incorporates five state-of-the-art somatic mutation callers, and extracts over 70 individual genomic and sequencing features for each candidate site. A training set is provided to an adaptively boosted decision tree learner to create a classifier for predicting mutation statuses. We validate our results with both synthetic and real data. We report that SomaticSeq is able to achieve better overall accuracy than any individual tool incorporated. Electronic supplementary material The online version of this article (doi:10.1186/s13059-015-0758-2) contains supplementary material, which is available to authorized users.