Feature-based classifiers for somatic mutation detection in tumour-normal paired sequencing data

Machine learning for the advancement of genome-scale metabolic modeling

Constraint-based modeling (CBM) has evolved as the core systems biology tool to map the interrelations between genotype, phenotype, and external environment. The recent advancement of high-throughput experimental approaches and multi-omics strategies has generated a plethora of new and precise information from wide-ranging biological domains. On the other hand, the continuously growing field of machine learning (ML) and its specialized branch of deep learning (DL) provide essential computational architectures for decoding complex and heterogeneous biological data. In recent years, both multi-omics and ML have assisted in the escalation of CBM. Condition-specific omics data, such as transcriptomics and proteomics, helped contextualize the model prediction while analyzing a particular phenotypic signature. At the same time, the advanced ML tools have eased the model reconstruction and analysis to increase the accuracy and prediction power. However, the development of these multi-disciplinary methodological frameworks mainly occurs independently, which limits the concatenation of biological knowledge from different domains. Hence, we have reviewed the potential of integrating multi-disciplinary tools and strategies from various fields, such as synthetic biology, CBM, omics, and ML, to explore the biochemical phenomenon beyond the conventional biological dogma. How the integrative knowledge of these intersected domains has improved bioengineering and biomedical applications has also been highlighted. We categorically explained the conventional genome-scale metabolic model (GEM) reconstruction tools and their improvement strategies through ML paradigms. Further, the crucial role of ML and DL in omics data restructuring for GEM development has also been briefly discussed. Finally, the case-study-based assessment of the state-of-the-art method for improving biomedical and metabolic engineering strategies has been elaborated. Therefore, this review demonstrates how integrating experimental and in silico strategies can help map the ever-expanding knowledge of biological systems driven by condition-specific cellular information. This multiview approach will elevate the application of ML-based CBM in the biomedical and bioengineering fields for the betterment of society and the environment.

AI-based histopathology image analysis reveals a distinct subset of endometrial cancers

Endometrial cancer (EC) has four molecular subtypes with strong prognostic value and therapeutic implications. The most common subtype (NSMP; No Specific Molecular Profile) is assigned after exclusion of the defining features of the other three molecular subtypes and includes patients with heterogeneous clinical outcomes. In this study, we employ artificial intelligence (AI)-powered histopathology image analysis to differentiate between p53abn and NSMP EC subtypes and consequently identify a sub-group of NSMP EC patients that has markedly inferior progression-free and disease-specific survival (termed 'p53abn-like NSMP'), in a discovery cohort of 368 patients and two independent validation cohorts of 290 and 614 from other centers. Shallow whole genome sequencing reveals a higher burden of copy number abnormalities in the 'p53abn-like NSMP' group compared to NSMP, suggesting that this group is biologically distinct compared to other NSMP ECs. Our work demonstrates the power of AI to detect prognostically different and otherwise unrecognizable subsets of EC where conventional and standard molecular or pathologic criteria fall short, refining image-based tumor classification. This study's findings are applicable exclusively to females.

Whole genome and transcriptome integrated analyses guide clinical care of pediatric poor prognosis cancers

The role for routine whole genome and transcriptome analysis (WGTA) for poor prognosis pediatric cancers remains undetermined. Here, we characterize somatic mutations, structural rearrangements, copy number variants, gene expression, immuno-profiles and germline cancer predisposition variants in children and adolescents with relapsed, refractory or poor prognosis malignancies who underwent somatic WGTA and matched germline sequencing. Seventy-nine participants with a median age at enrollment of 8.8 y (range 6 months to 21.2 y) are included. Germline pathogenic/likely pathogenic variants are identified in 12% of participants, of which 60% were not known prior. Therapeutically actionable variants are identified by targeted gene report and whole genome in 32% and 62% of participants, respectively, and increase to 96% after integrating transcriptome analyses. Thirty-two molecularly informed therapies are pursued in 28 participants with 54% achieving a clinical benefit rate; objective response or stable disease ≥6 months. Integrated WGTA identifies therapeutically actionable variants in almost all tumors and are directly translatable to clinical care of children with poor prognosis cancers.

Exploration of Germline Correlates and Risk of Immune-Related Adverse Events in Advanced Cancer Patients Treated with Immune Checkpoint Inhibitors

Immune checkpoint inhibitors (ICIs) are increasingly used in the treatment of many tumor types, and durable responses can be observed in select populations. However, patients may exhibit significant immune-related adverse events (irAEs) that may lead to morbidity. There is limited information on whether the presence of specific germline mutations may highlight those at elevated risk of irAEs. We evaluated 117 patients with metastatic solid tumors or hematologic malignancies who underwent genomic analysis through the ongoing Personalized OncoGenomics (POG) program at BC Cancer and received an ICI during their treatment history. Charts were reviewed for irAEs. Whole genome sequencing of a fresh biopsy and matched normal specimens (blood) was performed at the time of POG enrollment. Notably, we found that MHC class I alleles in the HLA-B27 family, which have been previously associated with autoimmune conditions, were associated with grade 3 hepatitis and pneumonitis (q = 0.007) in patients treated with combination PD-1/PD-L1 and CTLA-4 inhibitors, and PD-1 inhibitors in combination with IDO-1 inhibitors. These data highlight that some patients may have a genetic predisposition to developing irAEs.

What makes TMB an ambivalent biomarker for immunotherapy? A subtle mismatch between the sample-based design of variant callers and real clinical cohort

Tumor mutation burden (TMB) is a widely recognized biomarker for predicting the efficacy of immunotherapy. However, its use still remains highly controversial. In this study, we examine the underlying causes of this controversy based on clinical needs. By tracing the source of the TMB errors and analyzing the design philosophy behind variant callers, we identify the conflict between the incompleteness of biostatistics rules and the variety of clinical samples as the critical issue that renders TMB an ambivalent biomarker. A series of experiments were conducted to illustrate the challenges of mutation detection in clinical practice. Additionally, we also discuss potential strategies for overcoming these conflict issues to enable the application of TMB in guiding decision-making in real clinical settings.

Validation of genetic variants from NGS data using deep convolutional neural networks

Accurate somatic variant calling from next-generation sequencing data is one most important tasks in personalised cancer therapy. The sophistication of the available technologies is ever-increasing, yet, manual candidate refinement is still a necessary step in state-of-the-art processing pipelines. This limits reproducibility and introduces a bottleneck with respect to scalability. We demonstrate that the validation of genetic variants can be improved using a machine learning approach resting on a Convolutional Neural Network, trained using existing human annotation. In contrast to existing approaches, we introduce a way in which contextual data from sequencing tracks can be included into the automated assessment. A rigorous evaluation shows that the resulting model is robust and performs on par with trained researchers following published standard operating procedure.

Immune Activation following Irbesartan Treatment in a Colorectal Cancer Patient: A Case Study

Colorectal cancers are one of the most prevalent tumour types worldwide and, despite the emergence of targeted and biologic therapies, have among the highest mortality rates. The Personalized OncoGenomics (POG) program at BC Cancer performs whole genome and transcriptome analysis (WGTA) to identify specific alterations in an individual's cancer that may be most effectively targeted. Informed using WGTA, a patient with advanced mismatch repair-deficient colorectal cancer was treated with the antihypertensive drug irbesartan and experienced a profound and durable response. We describe the subsequent relapse of this patient and potential mechanisms of response using WGTA and multiplex immunohistochemistry (m-IHC) profiling of biopsies before and after treatment from the same metastatic site of the L3 spine. We did not observe marked differences in the genomic landscape before and after treatment. Analyses revealed an increase in immune signalling and infiltrating immune cells, particularly CD8+ T cells, in the relapsed tumour. These results indicate that the observed anti-tumour response to irbesartan may have been due to an activated immune response. Determining whether there may be other cancer contexts in which irbesartan may be similarly valuable will require additional studies.

Single-cell genomic variation induced by mutational processes in cancer

How cell-to-cell copy number alterations that underpin genomic instability1 in human cancers drive genomic and phenotypic variation, and consequently the evolution of cancer2, remains understudied. Here, by applying scaled single-cell whole-genome sequencing3 to wild-type, TP53-deficient and TP53-deficient;BRCA1-deficient or TP53-deficient;BRCA2-deficient mammary epithelial cells (13,818 genomes), and to primary triple-negative breast cancer (TNBC) and high-grade serous ovarian cancer (HGSC) cells (22,057 genomes), we identify three distinct 'foreground' mutational patterns that are defined by cell-to-cell structural variation. Cell- and clone-specific high-level amplifications, parallel haplotype-specific copy number alterations and copy number segment length variation (serrate structural variations) had measurable phenotypic and evolutionary consequences. In TNBC and HGSC, clone-specific high-level amplifications in known oncogenes were highly prevalent in tumours bearing fold-back inversions, relative to tumours with homologous recombination deficiency, and were associated with increased clone-to-clone phenotypic variation. Parallel haplotype-specific alterations were also commonly observed, leading to phylogenetic evolutionary diversity and clone-specific mono-allelic expression. Serrate variants were increased in tumours with fold-back inversions and were highly correlated with increased genomic diversity of cellular populations. Together, our findings show that cell-to-cell structural variation contributes to the origins of phenotypic and evolutionary diversity in TNBC and HGSC, and provide insight into the genomic and mutational states of individual cancer cells.

Accurate determination of CRISPR-mediated gene fitness in transplantable tumours

Assessing tumour gene fitness in physiologically-relevant model systems is challenging due to biological features of in vivo tumour regeneration, including extreme variations in single cell lineage progeny. Here we develop a reproducible, quantitative approach to pooled genetic perturbation in patient-derived xenografts (PDXs), by encoding single cell output from transplanted CRISPR-transduced cells in combination with a Bayesian hierarchical model. We apply this to 181 PDX transplants from 21 breast cancer patients. We show that uncertainty in fitness estimates depends critically on the number of transplant cell clones and the variability in clone sizes. We use a pathway-directed allelic series to characterize Notch signaling, and quantify TP53 / MDM2 drug-gene conditional fitness in outlier patients. We show that fitness outlier identification can be mirrored by pharmacological perturbation. Overall, we demonstrate that the gene fitness landscape in breast PDXs is dominated by inter-patient differences.

The impact of whole genome and transcriptome analysis (WGTA) on predictive biomarker discovery and diagnostic accuracy of advanced malignancies

In this study, we evaluate the impact of whole genome and transcriptome analysis (WGTA) on predictive molecular profiling and histologic diagnosis in a cohort of advanced malignancies. WGTA was used to generate reports including molecular alterations and site/tissue of origin prediction. Two reviewers analyzed genomic reports, clinical history, and tumor pathology. We used National Comprehensive Cancer Network (NCCN) consensus guidelines, Food and Drug Administration (FDA) approvals, and provincially reimbursed treatments to define genomic biomarkers associated with approved targeted therapeutic options (TTOs). Tumor tissue/site of origin was reassessed for most cases using genomic analysis, including a machine learning algorithm (Supervised Cancer Origin Prediction Using Expression [SCOPE]) trained on The Cancer Genome Atlas data. WGTA was performed on 652 cases, including a range of primary tumor types/tumor sites and 15 malignant tumors of uncertain histogenesis (MTUH). At the time WGTA was performed, alterations associated with an approved TTO were identified in 39 (6%) cases; 3 of these were not identified through routine pathology workup. In seven (1%) cases, the pathology workup either failed, was not performed, or gave a different result from the WGTA. Approved TTOs identified by WGTA increased to 103 (16%) when applying 2021 guidelines. The histopathologic diagnosis was reviewed in 389 cases and agreed with the diagnostic consensus after WGTA in 94% of non‐MTUH cases (n = 374). The remainder included situations where the morphologic diagnosis was changed based on WGTA and clinical data (0.5%), or where the WGTA was non‐contributory (5%). The 15 MTUH were all diagnosed as specific tumor types by WGTA. Tumor board reviews including WGTA agreed with almost all initial predictive molecular profile and histopathologic diagnoses. WGTA was a powerful tool to assign site/tissue of origin in MTUH. Current efforts focus on improving therapeutic predictive power and decreasing cost to enhance use of WGTA data as a routine clinical test.

Results of the phase I CCTG IND.231 trial of CX-5461 in patients with advanced solid tumors enriched for DNA-repair deficiencies

CX-5461 is a G-quadruplex stabilizer that exhibits synthetic lethality in homologous recombination-deficient models. In this multicentre phase I trial in patients with solid tumors, 40 patients are treated across 10 dose levels (50–650 mg/m²) to determine the recommended phase II dose (primary outcome), and evaluate safety, tolerability, pharmacokinetics (secondary outcomes). Defective homologous recombination is explored as a predictive biomarker of response. CX-5461 is generally well tolerated, with a recommended phase II dose of 475 mg/m² days 1, 8 and 15 every 4 weeks, and dose limiting phototoxicity. Responses are observed in 14% of patients, primarily in patients with defective homologous recombination. Reversion mutations in PALB2 and BRCA2 are detected on progression following initial response in germline carriers, confirming the underlying synthetic lethal mechanism. In vitro characterization of UV sensitization shows this toxicity is related to the CX-5461 chemotype, independent of G-quadruplex synthetic lethality. These results establish clinical proof-of-concept for this G-quadruplex stabilizer. Clinicaltrials.gov NCT02719977.

G-quadruplex stabilizers, including CX-5461, exhibit synthetic lethality with loss of BRCA1/2 in preclinical models. Here the authors report the results of a phase I study of CX-5461 in patients with solid tumors enriched for DNA-repair deficiencies.

Computational analysis of cancer genome sequencing data

Distilling biologically meaningful information from cancer genome sequencing data requires comprehensive identification of somatic alterations using rigorous computational methods. As the amount and complexity of sequencing data have increased, so has the number of tools for analysing them. Here, we describe the main steps involved in the bioinformatic analysis of cancer genomes, review key algorithmic developments and highlight popular tools and emerging technologies. These tools include those that identify point mutations, copy number alterations, structural variations and mutational signatures in cancer genomes. We also discuss issues in experimental design, the strengths and limitations of sequencing modalities and methodological challenges for the future.

Histologic and Genotypic Characterization of Lung Cancer in the Inuit Population of the Eastern Canadian Arctic

Inuit are the Indigenous Arctic peoples and residents of the Canadian territory of Nunavut who have the highest global rate of lung cancer. Given lung cancer’s mortality, histological and genomic characterization was undertaken to better understand the disease biology. We retrospectively studied all Inuit cases from Nunavut’s Qikiqtani (Baffin) region, referred to the Ottawa Hospital Cancer Center between 2001 and 2011. Demographics were compiled from medical records and tumor samples underwent pathologic/histologic confirmation. Tumors were analyzed by next generation sequencing (NGS) with a cancer hotspot mutation panel. Of 98 patients, the median age was 66 years and 61% were male. Tobacco use was reported in 87%, and 69% had a history of lung disease (tuberculosis or other). Histological types were: non-small cell lung carcinoma (NSCLC), 81%; small cell lung carcinoma, 16%. Squamous cell carcinoma (SCC) represented 65% of NSCLC. NGS on 55 samples demonstrated mutation rates similar to public lung cancer datasets. In SCC, the STK11 F354L mutation was observed at higher frequency than previously reported. This is the first study to characterize the histologic/genomic profiles of lung cancer in this population. A high incidence of SCC, and an elevated rate of STK11 mutations distinguishes this group from the North American population.

A platform for oncogenomic reporting and interpretation

Manual interpretation of variants remains rate limiting in precision oncology. The increasing scale and complexity of molecular data generated from comprehensive sequencing of cancer samples requires advanced interpretative platforms as precision oncology expands beyond individual patients to entire populations. To address this unmet need, we introduce a Platform for Oncogenomic Reporting and Interpretation (PORI), comprising an analytic framework that facilitates the interpretation and reporting of somatic variants in cancer. PORI integrates reporting and graph knowledge base tools combined with support for manual curation at the reporting stage. PORI represents an open-source platform alternative to commercial reporting solutions suitable for comprehensive genomic data sets in precision oncology. We demonstrate the utility of PORI by matching 9,961 pan-cancer genome atlas tumours to the graph knowledge base, calculating therapeutically informative alterations, and making available reports describing select individual samples.

The interpretation of somatic variants in cancer is challenging due to the scale and complexity of sequencing data. Here, the authors present PORI, an open-source framework for interpreting somatic variants in cancer using graph knowledge base tools, automated reporting, and manual curation.

Clinical response to nivolumab in an INI1-deficient pediatric chordoma correlates with immunogenic recognition of brachyury

Poorly differentiated chordoma (PDC) is a recently recognized subtype of chordoma characterized by expression of the embryonic transcription factor, brachyury, and loss of INI1. PDC primarily affects children and is associated with a poor prognosis and limited treatment options. Here we describe the molecular and immune tumour microenvironment profiles of two paediatric PDCs produced using whole-genome, transcriptome and whole-genome bisulfite sequencing (WGBS) and multiplex immunohistochemistry. Our analyses revealed the presence of tumour-associated immune cells, including CD8+ T cells, and expression of the immune checkpoint protein, PD-L1, in both patient samples. Molecular profiling provided the rationale for immune checkpoint inhibitor (ICI) therapy, which resulted in a clinical and radiographic response. A dominant T cell receptor (TCR) clone specific for a brachyury peptide-MHC complex was identified from bulk RNA sequencing, suggesting that targeting of the brachyury tumour antigen by tumour-associated T cells may underlie this clinical response to ICI. Correlative analysis with rhabdoid tumours, another INI1-deficient paediatric malignancy, suggests that a subset of tumours may share common immune phenotypes, indicating the potential for a therapeutically targetable subgroup of challenging paediatric cancers.

Exploratory study on classification of diabetes mellitus through a combined Random Forest Classifier

BACKGROUND

Diabetes Mellitus (DM) has become the third chronic non-communicable disease that hits patients after tumors, cardiovascular and cerebrovascular diseases, and has become one of the major public health problems in the world. Therefore, it is of great importance to identify individuals at high risk for DM in order to establish prevention strategies for DM.

METHODS

Aiming at the problem of high-dimensional feature space and high feature redundancy of medical data, as well as the problem of data imbalance often faced. This study explored different supervised classifiers, combined with SVM-SMOTE and two feature dimensionality reduction methods (Logistic stepwise regression and LAASO) to classify the diabetes survey sample data with unbalanced categories and complex related factors. Analysis and discussion of the classification results of 4 supervised classifiers based on 4 data processing methods. Five indicators including Accuracy, Precision, Recall, F1-Score and AUC are selected as the key indicators to evaluate the performance of the classification model.

RESULTS

According to the result, Random Forest Classifier combining SVM-SMOTE resampling technology and LASSO feature screening method (Accuracy = 0.890, Precision = 0.869, Recall = 0.919, F1-Score = 0.893, AUC = 0.948) proved the best way to tell those at high risk of DM. Besides, the combined algorithm helps enhance the classification performance for prediction of high-risk people of DM. Also, age, region, heart rate, hypertension, hyperlipidemia and BMI are the top six most critical characteristic variables affecting diabetes.

CONCLUSIONS

The Random Forest Classifier combining with SVM-SMOTE and LASSO feature reduction method perform best in identifying high-risk people of DM from individuals. And the combined method proposed in the study would be a good tool for early screening of DM.

Bioinformatics workflows for clinical applications in precision oncology

High-throughput molecular profiling of tumors is a fundamental aspect of precision oncology, enabling the identification of genomic alterations that can be targeted therapeutically. In this context, a patient is matched to a specific drug or therapy based on the tumor's underlying genetic driver events rather than the histologic classification. This approach requires extensive bioinformatics methodology and workflows, including raw sequencing data processing and quality control, variant calling and annotation, integration of different molecular data types, visualization and finally reporting the data to physicians, cancer researchers and pharmacologists in a format that is readily interpretable for clinical decision making. This review comprises a broad overview of these bioinformatics aspects and discusses the multiple analytical, technical and interpretational challenges that remain to efficiently translate molecular findings into personalized treatment recommendations.

An Efficient and Effective Model to Handle Missing Data in Classification

Missing data is one of the most important causes in reduction of classification accuracy. Many real datasets suffer from missing values, especially in medical sciences. Imputation is a common way to deal with incomplete datasets. There are various imputation methods that can be applied, and the choice of the best method depends on the dataset conditions such as sample size, missing percent, and missing mechanism. Therefore, the better solution is to classify incomplete datasets without imputation and without any loss of information. The structure of the “Bayesian additive regression trees” (BART) model is improved with the “Missingness Incorporated in Attributes” approach to solve its inefficiency in handling the missingness problem. Implementation of MIA-within-BART is named “BART.m”. As the abilities of BART.m are not investigated in classification of incomplete datasets, this simulation-based study aimed to provide such resource. The results indicate that BART.m can be used even for datasets with 90 missing present and more importantly, it diagnoses the irrelevant variables and removes them by its own. BART.m outperforms common models for classification with incomplete data, according to accuracy and computational time. Based on the revealed properties, it can be said that BART.m is a high accuracy model in classification of incomplete datasets which avoids any assumptions and preprocess steps.

Quantifying the influence of mutation detection on tumour subclonal reconstruction

Whole-genome sequencing can be used to estimate subclonal populations in tumours and this intra-tumoural heterogeneity is linked to clinical outcomes. Many algorithms have been developed for subclonal reconstruction, but their variabilities and consistencies are largely unknown. We evaluate sixteen pipelines for reconstructing the evolutionary histories of 293 localized prostate cancers from single samples, and eighteen pipelines for the reconstruction of 10 tumours with multi-region sampling. We show that predictions of subclonal architecture and timing of somatic mutations vary extensively across pipelines. Pipelines show consistent types of biases, with those incorporating SomaticSniper and Battenberg preferentially predicting homogenous cancer cell populations and those using MuTect tending to predict multiple populations of cancer cells. Subclonal reconstructions using multi-region sampling confirm that single-sample reconstructions systematically underestimate intra-tumoural heterogeneity, predicting on average fewer than half of the cancer cell populations identified by multi-region sequencing. Overall, these biases suggest caution in interpreting specific architectures and subclonal variants.

DeepSSV: detecting somatic small variants in paired tumor and normal sequencing data with convolutional neural network

It is of considerable interest to detect somatic mutations in paired tumor and normal sequencing data. A number of callers that are based on statistical or machine learning approaches have been developed to detect somatic small variants. However, they take into consideration only limited information about the reference and potential variant allele in both tumor and normal samples at a candidate somatic site. Also, they differ in how biological and technological noises are addressed. Hence, they are expected to produce divergent outputs. To overcome the drawbacks of existing somatic callers, we develop a deep learning-based tool called DeepSSV, which employs a convolutional neural network (CNN) model to learn increasingly abstract feature representations from the raw data in higher feature layers. DeepSSV creates a spatially oriented representation of read alignments around the candidate somatic sites adapted for the convolutional architecture, which enables it to expand to effectively gather scattered evidence. Moreover, DeepSSV incorporates the mapping information of both reference allele-supporting and variant allele-supporting reads in the tumor and normal samples at a genomic site that are readily available in the pileup format file. Together, the CNN model can process the whole alignment information. Such representational richness allows the model to capture the dependencies in the sequence and identify context-based sequencing artifacts. We fitted the model on ground truth somatic mutations and did benchmarking experiments on simulated and real tumors. The benchmarking results demonstrate that DeepSSV outperforms its state-of-the-art competitors in overall F1 score.

Analysis of Ugandan cervical carcinomas identifies human papillomavirus clade-specific epigenome and transcriptome landscapes

Cervical cancer is the most common cancer affecting sub-Saharan African women and is prevalent among HIV-positive (HIV+) individuals. No comprehensive profiling of cancer genomes, transcriptomes or epigenomes has been performed in this population thus far. We characterized 118 tumors from Ugandan patients, of whom 72 were HIV+, and performed extended mutation analysis on an additional 89 tumors. We detected human papillomavirus (HPV)-clade-specific differences in tumor DNA methylation, promoter- and enhancer-associated histone marks, gene expression and pathway dysregulation. Changes in histone modification at HPV integration events were correlated with upregulation of nearby genes and endogenous retroviruses.

SomaticCombiner: improving the performance of somatic variant calling based on evaluation tests and a consensus approach

It is challenging to identify somatic variants from high-throughput sequence reads due to tumor heterogeneity, sub-clonality, and sequencing artifacts. In this study, we evaluated the performance of eight primary somatic variant callers and multiple ensemble methods using both real and synthetic whole-genome sequencing, whole-exome sequencing, and deep targeted sequencing datasets with the NA12878 cell line. The test results showed that a simple consensus approach can significantly improve performance even with a limited number of callers and is more robust and stable than machine learning based ensemble approaches. To fully exploit the multi-callers, we also developed a software package, SomaticCombiner, that can combine multiple callers and integrates a new variant allelic frequency (VAF) adaptive majority voting approach, which can maintain sensitive detection for variants with low VAFs.

SMuRF: portable and accurate ensemble prediction of somatic mutations

Summary

Somatic Mutation calling method using a Random Forest (SMuRF) integrates predictions and auxiliary features from multiple somatic mutation callers using a supervised machine learning approach. SMuRF is trained on community-curated matched tumor and normal whole genome sequencing data. SMuRF predicts both SNVs and indels with high accuracy in genome or exome-level sequencing data. Furthermore, the method is robust across multiple tested cancer types and predicts low allele frequency variants with high accuracy. In contrast to existing ensemble-based somatic mutation calling approaches, SMuRF works out-of-the-box and is orders of magnitudes faster.

Availability and implementation

The method is implemented in R and available at https://github.com/skandlab/SMuRF. SMuRF operates as an add-on to the community-developed bcbio-nextgen somatic variant calling pipeline.

Supplementary information

Supplementary data are available at Bioinformatics online.

PRESM: personalized reference editor for somatic mutation discovery in cancer genomics

MOTIVATION

Accurate detection of somatic mutations is a crucial step toward understanding cancer. Various tools have been developed to detect somatic mutations from cancer genome sequencing data by mapping reads to a universal reference genome and inferring likelihoods from complex statistical models. However, read mapping is frequently obstructed by mismatches between germline and somatic mutations on a read and the reference genome. Previous attempts to develop personalized genome tools are not compatible with downstream statistical models for somatic mutation detection.

RESULTS

We present PRESM, a tool that builds personalized reference genomes by integrating germline mutations into the reference genome. The aforementioned obstacle is circumvented by using a two-step germline substitution procedure, maintaining positional fidelity using an innovative workaround. Reads derived from tumor tissue can be positioned more accurately along a personalized reference than a universal reference due to the reduced genetic distance between the subject (tumor genome) and the target (the personalized genome). Application of PRESM's personalized genome reduced false-positive (FP) somatic mutation calls by as much as 55.5%, and facilitated the discovery of a novel somatic point mutation on a germline insertion in PDE1A, a phosphodiesterase associated with melanoma. Moreover, all improvements in calling accuracy were achieved without parameter optimization, as PRESM itself is parameter-free. Hence, similar increases in read mapping and decreases in the FP rate will persist when PRESM-built genomes are applied to any user-provided dataset.

AVAILABILITY AND IMPLEMENTATION

The software is available at https://github.com/precisionomics/PRESM.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

The promise of toxicogenomics for genetic toxicology: past, present and future

Toxicogenomics, the application of genomics to toxicology, was described as 'a new era' for toxicology. Standard toxicity tests typically involve a number of short-term bioassays that are costly, time consuming, require large numbers of animals and generally focus on a single end point. Toxicogenomics was heralded as a way to improve the efficiency of toxicity testing by assessing gene regulation across the genome, allowing rapid classification of compounds based on characteristic expression profiles. Gene expression microarrays could measure and characterise genome-wide gene expression changes in a single study and while transcriptomic profiles that can discriminate between genotoxic and non-genotoxic carcinogens have been identified, challenges with the approach limited its application. As such, toxicogenomics did not transform the field of genetic toxicology in the way it was predicted. More recently, next generation sequencing (NGS) technologies have revolutionised genomics owing to the fact that hundreds of billions of base pairs can be sequenced simultaneously cheaper and quicker than traditional Sanger methods. In relation to genetic toxicology, and thousands of cancer genomes have been sequenced with single-base substitution mutational signatures identified, and mutation signatures have been identified following treatment of cells with known or suspected environmental carcinogens. RNAseq has been applied to detect transcriptional changes following treatment with genotoxins; modified RNAseq protocols have been developed to identify adducts in the genome and Duplex sequencing is an example of a technique that has recently been developed to accurately detect mutation. Machine learning, including MutationSeq and SomaticSeq, has also been applied to somatic mutation detection and improvements in automation and/or the application of machine learning algorithms may allow high-throughput mutation sequencing in the future. This review will discuss the initial promise of transcriptomics for genetic toxicology, and how the development of NGS technologies and new machine learning algorithms may finally realise that promise.

BATCAVE: calling somatic mutations with a tumor- and site-specific prior

Detecting somatic mutations withins tumors is key to understanding treatment resistance, patient prognosis and tumor evolution. Mutations at low allelic frequency, those present in only a small portion of tumor cells, are particularly difficult to detect. Many algorithms have been developed to detect such mutations, but none models a key aspect of tumor biology. Namely, every tumor has its own profile of mutation types that it tends to generate. We present BATCAVE (Bayesian Analysis Tools for Context-Aware Variant Evaluation), an algorithm that first learns the individual tumor mutational profile and mutation rate then uses them in a prior for evaluating potential mutations. We also present an R implementation of the algorithm, built on the popular caller MuTect. Using simulations, we show that adding the BATCAVE algorithm to MuTect improves variant detection. It also improves the calibration of posterior probabilities, enabling more principled tradeoff between precision and recall. We also show that BATCAVE performs well on real data. Our implementation is computationally inexpensive and straightforward to incorporate into existing MuTect pipelines. More broadly, the algorithm can be added to other variant callers, and it can be extended to include additional biological features that affect mutation generation.

A community effort to create standards for evaluating tumor subclonal reconstruction

Tumor DNA sequencing data can be interpreted by computational methods that analyze genomic heterogeneity to infer evolutionary dynamics. A growing number of studies have used these approaches to link cancer evolution with clinical progression and response to therapy. Although the inference of tumor phylogenies is rapidly becoming standard practice in cancer genome analyses, standards for evaluating them are lacking. To address this need, we systematically assess methods for reconstructing tumor subclonality. First, we elucidate the main algorithmic problems in subclonal reconstruction and develop quantitative metrics for evaluating them. Then we simulate realistic tumor genomes that harbor all known clonal and subclonal mutation types and processes. Finally, we benchmark 580 tumor reconstructions, varying tumor read depth, tumor type and somatic variant detection. Our analysis provides a baseline for the establishment of gold-standard methods to analyze tumor heterogeneity.

Identification and Analyses of Extra-Cranial and Cranial Rhabdoid Tumor Molecular Subgroups Reveal Tumors with Cytotoxic T Cell Infiltration

Extra-cranial malignant rhabdoid tumors (MRTs) and cranial atypical teratoid RTs (ATRTs) are heterogeneous pediatric cancers driven primarily by SMARCB1 loss. To understand the genome-wide molecular relationships between MRTs and ATRTs, we analyze multi-omics data from 140 MRTs and 161 ATRTs. We detect similarities between the MYC subgroup of ATRTs (ATRT-MYC) and extra-cranial MRTs, including global DNA hypomethylation and overexpression of HOX genes and genes involved in mesenchymal development, distinguishing them from other ATRT subgroups that express neural-like features. We identify five DNA methylation subgroups associated with anatomical sites and SMARCB1 mutation patterns. Groups 1, 3, and 4 exhibit cytotoxic T cell infiltration and expression of immune checkpoint regulators, consistent with a potential role for immunotherapy in rhabdoid tumor patients.

Clonal Decomposition and DNA Replication States Defined by Scaled Single-Cell Genome Sequencing

Accurate measurement of clonal genotypes, mutational processes, and replication states from individual tumor-cell genomes will facilitate improved understanding of tumor evolution. We have developed DLP+, a scalable single-cell whole-genome sequencing platform implemented using commodity instruments, image-based object recognition, and open source computational methods. Using DLP+, we have generated a resource of 51,926 single-cell genomes and matched cell images from diverse cell types including cell lines, xenografts, and diagnostic samples with limited material. From this resource we have defined variation in mitotic mis-segregation rates across tissue types and genotypes. Analysis of matched genomic and image measurements revealed correlations between cellular morphology and genome ploidy states. Aggregation of cells sharing copy number profiles allowed for calculation of single-nucleotide resolution clonal genotypes and inference of clonal phylogenies and avoided the limitations of bulk deconvolution. Finally, joint analysis over the above features defined clone-specific chromosomal aneuploidy in polyclonal populations.

Bayesian additive regression trees and the General BART model

Bayesian additive regression trees (BART) is a flexible prediction model/machine learning approach that has gained widespread popularity in recent years. As BART becomes more mainstream, there is an increased need for a paper that walks readers through the details of BART, from what it is to why it works. This tutorial is aimed at providing such a resource. In addition to explaining the different components of BART using simple examples, we also discuss a framework, the General BART model that unifies some of the recent BART extensions, including semiparametric models, correlated outcomes, and statistical matching problems in surveys, and models with weaker distributional assumptions. By showing how these models fit into a single framework, we hope to demonstrate a simple way of applying BART to research problems that go beyond the original independent continuous or binary outcomes framework.

Somatic mutations - Evolution within the individual

With the rapid advancement of sequencing technologies over the last two decades, it is becoming feasible to detect rare variants from somatic tissue samples. Studying such somatic mutations can provide deep insights into various senescence-related diseases, including cancer, inflammation, and sporadic psychiatric disorders. While it is still a difficult task to identify true somatic mutations, relentless efforts to combine experimental and computational methods have made it possible to obtain reliable data. Furthermore, state-of-the-art machine learning approaches have drastically improved the efficiency and sensitivity of these methods. Meanwhile, we can regard somatic mutations as a counterpart of germline mutations, and it is possible to apply well-formulated mathematical frameworks developed for population genetics and molecular evolution to analyze this 'somatic evolution'. For example, retrospective cell lineage tracing is a promising technique to elucidate the mechanism of pre-diseases using single-cell RNA-sequencing (scRNA-seq) data.

Comprehensive Outline of Whole Exome Sequencing Data Analysis Tools Available in Clinical Oncology

Whole exome sequencing (WES) enables the analysis of all protein coding sequences in the human genome. This technology enables the investigation of cancer-related genetic aberrations that are predominantly located in the exonic regions. WES delivers high-throughput results at a reasonable price. Here, we review analysis tools enabling utilization of WES data in clinical and research settings. Technically, WES initially allows the detection of single nucleotide variants (SNVs) and copy number variations (CNVs), and data obtained through these methods can be combined and further utilized. Variant calling algorithms for SNVs range from standalone tools to machine learning-based combined pipelines. Tools for CNV detection compare the number of reads aligned to a dedicated segment. Both SNVs and CNVs help to identify mutations resulting in pharmacologically druggable alterations. The identification of homologous recombination deficiency enables the use of PARP inhibitors. Determining microsatellite instability and tumor mutation burden helps to select patients eligible for immunotherapy. To pave the way for clinical applications, we have to recognize some limitations of WES, including its restricted ability to detect CNVs, low coverage compared to targeted sequencing, and the missing consensus regarding references and minimal application requirements. Recently, Galaxy became the leading platform in non-command line-based WES data processing. The maturation of next-generation sequencing is reinforced by Food and Drug Administration (FDA)-approved methods for cancer screening, detection, and follow-up. WES is on the verge of becoming an affordable and sufficiently evolved technology for everyday clinical use.

A machine learning approach for somatic mutation discovery

Variability in the accuracy of somatic mutation detection may affect the discovery of alterations and the therapeutic management of cancer patients. To address this issue, we developed a somatic mutation discovery approach based on machine learning that outperformed existing methods in identifying experimentally validated tumor alterations (sensitivity of 97% versus 90 to 99%; positive predictive value of 98% versus 34 to 92%). Analysis of paired tumor-normal exome data from 1368 TCGA (The Cancer Genome Atlas) samples using this method revealed concordance for 74% of mutation calls but also identified likely false-positive and false-negative changes in TCGA data, including in clinically actionable genes. Determination of high-quality somatic mutation calls improved tumor mutation load-based predictions of clinical outcome for melanoma and lung cancer patients previously treated with immune checkpoint inhibitors. Integration of high-quality machine learning mutation detection in clinical next-generation sequencing (NGS) analyses increased the accuracy of test results compared to other clinical sequencing analyses. These analyses provide an approach for improved identification of tumor-specific mutations and have important implications for research and clinical management of cancer patients.

Comprehensive genomic profiling of glioblastoma tumors, BTICs, and xenografts reveals stability and adaptation to growth environments

Glioblastoma multiforme (GBM) is the most deadly brain tumor, and currently lacks effective treatment options. Brain tumor-initiating cells (BTICs) and orthotopic xenografts are widely used in investigating GBM biology and new therapies for this aggressive disease. However, the genomic characteristics and molecular resemblance of these models to GBM tumors remain undetermined. We used massively parallel sequencing technology to decode the genomes and transcriptomes of BTICs and xenografts and their matched tumors in order to delineate the potential impacts of the distinct growth environments. Using data generated from whole-genome sequencing of 201 samples and RNA sequencing of 118 samples, we show that BTICs and xenografts resemble their parental tumor at the genomic level but differ at the mRNA expression and epigenomic levels, likely due to the different growth environment for each sample type. These findings suggest that a comprehensive genomic understanding of in vitro and in vivo GBM model systems is crucial for interpreting data from drug screens, and can help control for biases introduced by cell-culture conditions and the microenvironment in mouse models. We also found that lack of MGMT expression in pretreated GBM is linked to hypermutation, which in turn contributes to increased genomic heterogeneity and requires new strategies for GBM treatment.

Integrative genomic analysis of matched primary and metastatic pediatric osteosarcoma

Despite being the most common childhood bone tumor, the genomic characterization of osteosarcoma remains incomplete. In particular, very few osteosarcoma metastases have been sequenced to date, critical to better understand mechanisms of progression and evolution in this tumor. We performed an integrated whole genome and exome sequencing analysis of paired primary and metastatic pediatric osteosarcoma specimens to identify recurrent genomic alterations. Sequencing of 13 osteosarcoma patients including 13 primary, 10 metastatic, and 3 locally recurring tumors revealed a highly heterogeneous mutational landscape, including cases of hypermutation and microsatellite instability positivity, but with virtually no recurrent alterations except for mutations involving the tumor suppressor genes RB1 and TP53. At the germline level, we detected alterations in multiple cancer related genes in the majority of the cohort, including those potentially disrupting DNA damage response pathways. Metastases retained only a minimal number of short variants from their corresponding primary tumors, while copy number alterations showed higher conservation. One recurrently amplified gene, KDR, was highly expressed in advanced cases and associated with poor prognosis. © 2019 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Genomic characterization of a well-differentiated grade 3 pancreatic neuroendocrine tumor

Pancreatic neuroendocrine neoplasms (PanNENs) represent a minority of pancreatic neoplasms that exhibit variability in prognosis. Ongoing mutational analyses of PanNENs have found recurrent abnormalities in chromatin remodeling genes (e.g., DAXX and ATRX), and mTOR pathway genes (e.g., TSC2, PTEN PIK3CA, and MEN1), some of which have relevance to patients with related familial syndromes. Most recently, grade 3 PanNENs have been divided into two groups based on differentiation, creating a new group of well-differentiated grade 3 neuroendocrine tumors (PanNETs) that have had a limited whole-genome level characterization to date. In a patient with a metastatic well-differentiated grade 3 PanNET, our study utilized whole-genome sequencing of liver metastases for the comparative analysis and detection of single-nucleotide variants, insertions and deletions, structural variants, and copy-number variants, with their biologic relevance confirmed by RNA sequencing. We found that this tumor most notably exhibited a TSC1-disrupting fusion, showed a novel CHD7–BEND2 fusion, and lacked any somatic variants in ATRX, DAXX, and MEN1.

NeoMutate: an ensemble machine learning framework for the prediction of somatic mutations in cancer

Background

The accurate screening of tumor genomic landscapes for somatic mutations using high-throughput sequencing involves a crucial step in precise clinical diagnosis and targeted therapy. However, the complex inherent features of cancer tissue, especially, tumor genetic intra-heterogeneity coupled with the problem of sequencing and alignment artifacts, makes somatic variant calling a challenging task. Current variant filtering strategies, such as rule-based filtering and consensus voting of different algorithms, have previously helped to increase specificity, although comes at the cost of sensitivity.

Methods

In light of this, we have developed the NeoMutate framework which incorporates 7 supervised machine learning (ML) algorithms to exploit the strengths of multiple variant callers, using a non-redundant set of biological and sequence features. We benchmarked NeoMutate by simulating more than 10,000 bona fide cancer-related mutations into three well-characterized Genome in a Bottle (GIAB) reference samples.

Results

A robust and exhaustive evaluation of NeoMutate’s performance based on 5-fold cross validation experiments, in addition to 3 independent tests, demonstrated a substantially improved variant detection accuracy compared to any of its individual composite variant callers and consensus calling of multiple tools.

Conclusions

We show here that integrating multiple tools in an ensemble ML layer optimizes somatic variant detection rates, leading to a potentially improved variant selection framework for the diagnosis and treatment of cancer.

Electronic supplementary material

The online version of this article (10.1186/s12920-019-0508-5) contains supplementary material, which is available to authorized users.

NRG1 Gene Fusions Are Recurrent, Clinically Actionable Gene Rearrangements in KRAS Wild-Type Pancreatic Ductal Adenocarcinoma

PURPOSE

Gene fusions involving neuregulin 1 (NRG1) have been noted in multiple cancer types and have potential therapeutic implications. Although varying results have been reported in other cancer types, the efficacy of the HER-family kinase inhibitor afatinib in the treatment of NRG1 fusion-positive pancreatic ductal adenocarcinoma is not fully understood.

EXPERIMENTAL DESIGN

Forty-seven patients with pancreatic ductal adenocarcinoma received comprehensive whole-genome and transcriptome sequencing and analysis. Two patients with gene fusions involving NRG1 received afatinib treatment, with response measured by pretreatment and posttreatment PET/CT imaging.

RESULTS

Three of 47 (6%) patients with advanced pancreatic ductal adenocarcinoma were identified as KRAS wild type by whole-genome sequencing. All KRAS wild-type tumors were positive for gene fusions involving the ERBB3 ligand NRG1. Two of 3 patients with NRG1 fusion-positive tumors were treated with afatinib and demonstrated a significant and rapid response while on therapy.

CONCLUSIONS

This work adds to a growing body of evidence that NRG1 gene fusions are recurrent, therapeutically actionable genomic events in pancreatic cancers. Based on the clinical outcomes described here, patients with KRAS wild-type tumors harboring NRG1 gene fusions may benefit from treatment with afatinib.See related commentary by Aguirre, p. 4589.

Calling Variants in the Clinic: Informed Variant Calling Decisions Based on Biological, Clinical, and Laboratory Variables

Deep sequencing genomic analysis is becoming increasingly common in clinical research and practice, enabling accurate identification of diagnostic, prognostic, and predictive determinants. Variant calling, distinguishing between true mutations and experimental errors, is a central task of genomic analysis and often requires sophisticated statistical, computational, and/or heuristic techniques. Although variant callers seek to overcome noise inherent in biological experiments, variant calling can be significantly affected by outside factors including those used to prepare, store, and analyze samples. The goal of this review is to discuss known experimental features, such as sample preparation, library preparation, and sequencing, alongside diverse biological and clinical variables, and evaluate their effect on variant caller selection and optimization.

NONSTATIONARY SPATIAL PREDICTION OF SOIL ORGANIC CARBON: IMPLICATIONS FOR STOCK ASSESSMENT DECISION MAKING

The Rapid Carbon Assessment (RaCA) project was conducted by the US Department of Agriculture's National Resources Conservation Service between 2010-2012 in order to provide contemporaneous measurements of soil organic carbon (SOC) across the US. Despite the broad extent of the RaCA data collection effort, direct observations of SOC are not available at the high spatial resolution needed for studying carbon storage in soil and its implications for important problems in climate science and agriculture. As a result, there is a need for predicting SOC at spatial locations not included as part of the RaCA project. In this paper, we compare spatial prediction of SOC using a subset of the RaCA data for a variety of statistical methods. We investigate the performance of methods with off-the-shelf software available (both stationary and nonstationary) as well as a novel nonstationary approach based on partitioning relevant spatially-varying covariate processes. Our new method addresses open questions regarding (1) how to partition the spatial domain for segmentation-based nonstationary methods, (2) incorporating partially observed covariates into a spatial model, and (3) accounting for uncertainty in the partitioning. In applying the various statistical methods we find that there are minimal differences in out-of-sample criteria for this particular data set, however, there are major differences in maps of uncertainty in SOC predictions. We argue that the spatially-varying measures of prediction uncertainty produced by our new approach are valuable to decision makers, as they can be used to better benchmark mechanistic models, identify target areas for soil restoration projects, and inform carbon sequestration projects.

Will machine learning end the viability of radiology as a thriving medical specialty?

There have been tremendous advances in artificial intelligence (AI) and machine learning (ML) within the past decade, especially in the application of deep learning to various challenges. These include advanced competitive games (such as Chess and Go), self-driving cars, speech recognition, and intelligent personal assistants. Rapid advances in computer vision for recognition of objects in pictures have led some individuals, including computer science experts and health care system experts in machine learning, to make predictions that ML algorithms will soon lead to the replacement of the radiologist. However, there are complex technological, regulatory, and medicolegal obstacles facing the implementation of machine learning in radiology that will definitely preclude replacement of the radiologist by these algorithms within the next two decades and beyond. While not a comprehensive review of machine learning, this article is intended to highlight specific features of machine learning which face significant technological and health care systems challenges. Rather than replacing radiologists, machine learning will provide quantitative tools that will increase the value of diagnostic imaging as a biomarker, increase image quality with decreased acquisition times, and improve workflow, communication, and patient safety. In the foreseeable future, we predict that today's generation of radiologists will be replaced not by ML algorithms, but by a new breed of data science-savvy radiologists who have embraced and harnessed the incredible potential that machine learning has to advance our ability to care for our patients. In this way, radiology will remain a viable medical specialty for years to come.

Translating cancer genomics into precision medicine with artificial intelligence: applications, challenges and future perspectives

In the field of cancer genomics, the broad availability of genetic information offered by next-generation sequencing technologies and rapid growth in biomedical publication has led to the advent of the big-data era. Integration of artificial intelligence (AI) approaches such as machine learning, deep learning, and natural language processing (NLP) to tackle the challenges of scalability and high dimensionality of data and to transform big data into clinically actionable knowledge is expanding and becoming the foundation of precision medicine. In this paper, we review the current status and future directions of AI application in cancer genomics within the context of workflows to integrate genomic analysis for precision cancer care. The existing solutions of AI and their limitations in cancer genetic testing and diagnostics such as variant calling and interpretation are critically analyzed. Publicly available tools or algorithms for key NLP technologies in the literature mining for evidence-based clinical recommendations are reviewed and compared. In addition, the present paper highlights the challenges to AI adoption in digital healthcare with regard to data requirements, algorithmic transparency, reproducibility, and real-world assessment, and discusses the importance of preparing patients and physicians for modern digitized healthcare. We believe that AI will remain the main driver to healthcare transformation toward precision medicine, yet the unprecedented challenges posed should be addressed to ensure safety and beneficial impact to healthcare.

Somatic mutation detection and classification through probabilistic integration of clonal population information

Somatic mutations are a primary contributor to malignancy in human cells. Accurate detection of mutations is needed to define the clonal composition of tumours whereby clones may have distinct phenotypic properties. Although analysis of mutations over multiple tumour samples from the same patient has the potential to enhance identification of clones, few analytic methods exploit the correlation structure across samples. We posited that incorporating clonal information into joint analysis over multiple samples would improve mutation detection, particularly those with low prevalence. In this paper, we develop a new procedure called MuClone, for detection of mutations across multiple tumour samples of a patient from whole genome or exome sequencing data. In addition to mutation detection, MuClone classifies mutations into biologically meaningful groups and allows us to study clonal dynamics. We show that, on lung and ovarian cancer datasets, MuClone improves somatic mutation detection sensitivity over competing approaches without compromising specificity.

Whole-genome and transcriptome profiling of a metastatic thyroid-like follicular renal cell carcinoma

Thyroid-like follicular renal cell carcinoma (TLFRCC) is a rare cancer with few reports of metastatic disease. Little is known regarding genomic characteristics and therapeutic targets. We present the clinical, pathologic, genomic, and transcriptomic analyses of a case of a 27-yr-old male with TLFRCC who presented initially with bone metastases of unknown primary. Genomic DNA from peripheral blood and metastatic tumor samples were sequenced. A transcriptome of 280 million sequence reads was generated from the same tumor sample. Tumor somatic expression profiles were analyzed to detect aberrant expression. Genomic and transcriptomic data sets were integrated to reveal dysregulation in pathways and identify potential therapeutic targets. Integrative genomic analysis with The Cancer Genome Atlas (TCGA) data set revealed the following outliers in gene expression profiles: CDK6 (81st percentile), MYC (99th percentile), AR (100th percentile), PDGFRA and PDGFRB (99th and 100th percentiles, respectively), and MAP2K2 (86th percentile). The patient received first-line sunitinib to target PDGFRA and PDGFRB and had stable disease for >6 mo, followed by nivolumab upon progression. To the authors’ knowledge, this is the first reported case of comprehensive somatic genomic analyses in a patient with metastatic TLFRCC. Somatic analyses provided molecular confirmation of the primary site of cancer and potential therapeutic strategies in a rare disease with little evidence of efficacy on systemic therapy.

A deep learning approach to automate refinement of somatic variant calling from cancer sequencing data

Cancer genomic analysis requires accurate identification of somatic variants in sequencing data. Manual review to refine somatic variant calls is required as a final step after automated processing. However, manual variant refinement is time-consuming, costly, poorly standardized, and non-reproducible. Here, we systematized and standardized somatic variant refinement using a machine learning approach. The final model incorporates 41,000 variants from 440 sequencing cases. This model accurately recapitulated manual refinement labels for three independent testing sets (13,579 variants) and accurately predicted somatic variants confirmed by orthogonal validation sequencing data (212,158 variants). The model improves on manual somatic refinement by reducing bias on calls otherwise subject to high inter-reviewer variability.

Detection and benchmarking of somatic mutations in cancer genomes using RNA-seq data

To detect functional somatic mutations in tumor samples, whole-exome sequencing (WES) is often used for its reliability and relative low cost. RNA-seq, while generally used to measure gene expression, can potentially also be used for identification of somatic mutations. However there has been little systematic evaluation of the utility of RNA-seq for identifying somatic mutations. Here, we develop and evaluate a pipeline for processing RNA-seq data from glioblastoma multiforme (GBM) tumors in order to identify somatic mutations. The pipeline entails the use of the STAR aligner 2-pass procedure jointly with MuTect2 from genome analysis toolkit (GATK) to detect somatic variants. Variants identified from RNA-seq data were evaluated by comparison against the COSMIC and dbSNP databases, and also compared to somatic variants identified by exome sequencing. We also estimated the putative functional impact of coding variants in the most frequently mutated genes in GBM. Interestingly, variants identified by RNA-seq alone showed better representation of GBM-related mutations cataloged by COSMIC. RNA-seq-only data substantially outperformed the ability of WES to reveal potentially new somatic mutations in known GBM-related pathways, and allowed us to build a high-quality set of somatic mutations common to exome and RNA-seq calls. Using RNA-seq data in parallel with WES data to detect somatic mutations in cancer genomes can thus broaden the scope of discoveries and lend additional support to somatic variants identified by exome sequencing alone.