Prediction of the functional impact of missense variants in BRCA1 and BRCA2 with BRCA-ML

Variant Impact Predictor database (VIPdb), version 2: trends from three decades of genetic variant impact predictors

BACKGROUND

Variant interpretation is essential for identifying patients' disease-causing genetic variants amongst the millions detected in their genomes. Hundreds of Variant Impact Predictors (VIPs), also known as Variant Effect Predictors (VEPs), have been developed for this purpose, with a variety of methodologies and goals. To facilitate the exploration of available VIP options, we have created the Variant Impact Predictor database (VIPdb).

RESULTS

The Variant Impact Predictor database (VIPdb) version 2 presents a collection of VIPs developed over the past three decades, summarizing their characteristics, ClinGen calibrated scores, CAGI assessment results, publication details, access information, and citation patterns. We previously summarized 217 VIPs and their features in VIPdb in 2019. Building upon this foundation, we identified and categorized an additional 190 VIPs, resulting in a total of 407 VIPs in VIPdb version 2. The majority of the VIPs have the capacity to predict the impacts of single nucleotide variants and nonsynonymous variants. More VIPs tailored to predict the impacts of insertions and deletions have been developed since the 2010s. In contrast, relatively few VIPs are dedicated to the prediction of splicing, structural, synonymous, and regulatory variants. The increasing rate of citations to VIPs reflects the ongoing growth in their use, and the evolving trends in citations reveal development in the field and individual methods.

CONCLUSIONS

VIPdb version 2 summarizes 407 VIPs and their features, potentially facilitating VIP exploration for various variant interpretation applications. VIPdb is available at  https://genomeinterpretation.org/vipdb.

Variant Impact Predictor database (VIPdb), version 2: Trends from 25 years of genetic variant impact predictors

Background

Variant interpretation is essential for identifying patients' disease-causing genetic variants amongst the millions detected in their genomes. Hundreds of Variant Impact Predictors (VIPs), also known as Variant Effect Predictors (VEPs), have been developed for this purpose, with a variety of methodologies and goals. To facilitate the exploration of available VIP options, we have created the Variant Impact Predictor database (VIPdb).

Results

The Variant Impact Predictor database (VIPdb) version 2 presents a collection of VIPs developed over the past 25 years, summarizing their characteristics, ClinGen calibrated scores, CAGI assessment results, publication details, access information, and citation patterns. We previously summarized 217 VIPs and their features in VIPdb in 2019. Building upon this foundation, we identified and categorized an additional 186 VIPs, resulting in a total of 403 VIPs in VIPdb version 2. The majority of the VIPs have the capacity to predict the impacts of single nucleotide variants and nonsynonymous variants. More VIPs tailored to predict the impacts of insertions and deletions have been developed since the 2010s. In contrast, relatively few VIPs are dedicated to the prediction of splicing, structural, synonymous, and regulatory variants. The increasing rate of citations to VIPs reflects the ongoing growth in their use, and the evolving trends in citations reveal development in the field and individual methods.

Conclusions

VIPdb version 2 summarizes 403 VIPs and their features, potentially facilitating VIP exploration for various variant interpretation applications.

Availability

VIPdb version 2 is available at https://genomeinterpretation.org/vipdb.

Functional analysis and clinical classification of 462 germline BRCA2 missense variants affecting the DNA binding domain

Variants of uncertain significance (VUSs) in BRCA2 are a common result of hereditary cancer genetic testing. While more than 4,000 unique VUSs, comprised of missense or intronic variants, have been identified in BRCA2, the few missense variants now classified clinically as pathogenic or likely pathogenic are predominantly located in the region encoding the C-terminal DNA binding domain (DBD). We report on functional evaluation of the influence of 462 BRCA2 missense variants affecting the DBD on DNA repair activity of BRCA2 using a homology-directed DNA double-strand break repair assay. Of these, 137 were functionally abnormal, 313 were functionally normal, and 12 demonstrated intermediate function. Comparisons with other functional studies of BRCA2 missense variants yielded strong correlations. Sequence-based in silico prediction models had high sensitivity, but limited specificity, relative to the homology-directed repair assay. Combining the functional results with clinical and genetic data in an American College of Medical Genetics (ACMG)/Association for Molecular Pathology (AMP)-like variant classification framework from a clinical testing laboratory, after excluding known splicing variants and functionally intermediate variants, classified 431 of 442 (97.5%) missense variants (129 as pathogenic/likely pathogenic and 302 as benign/likely benign). Functionally abnormal variants classified as pathogenic by ACMG/AMP rules were associated with a slightly lower risk of breast cancer (odds ratio [OR] 5.15, 95% confidence interval [CI] 3.43-7.83) than BRCA2 DBD protein truncating variants (OR 8.56, 95% CI 6.03-12.36). Overall, functional studies of BRCA2 variants using validated assays substantially improved the variant classification yield from ACMG/AMP models and are expected to improve clinical management of many individuals found to harbor germline BRCA2 missense VUS.

Saturation genome editing of 11 codons and exon 13 of BRCA2 coupled with chemotherapeutic drug response accurately determines pathogenicity of variants

The unknown pathogenicity of a significant number of variants found in cancer-related genes is attributed to limited epidemiological data, resulting in their classification as variant of uncertain significance (VUS). To date, Breast Cancer gene-2 (BRCA2) has the highest number of VUSs, which has necessitated the development of several robust functional assays to determine their functional significance. Here we report the use of a humanized-mouse embryonic stem cell (mESC) line expressing a single copy of the human BRCA2 for a CRISPR-Cas9-based high-throughput functional assay. As a proof-of-principle, we have saturated 11 codons encoded by BRCA2 exons 3, 18, 19 and all possible single-nucleotide variants in exon 13 and multiplexed these variants for their functional categorization. Specifically, we used a pool of 180-mer single-stranded donor DNA to generate all possible combination of variants. Using a high throughput sequencing-based approach, we show a significant drop in the frequency of non-functional variants, whereas functional variants are enriched in the pool of the cells. We further demonstrate the response of these variants to the DNA-damaging agents, cisplatin and olaparib, allowing us to use cellular survival and drug response as parameters for variant classification. Using this approach, we have categorized 599 BRCA2 variants including 93-single nucleotide variants (SNVs) across the 11 codons, of which 28 are reported in ClinVar. We also functionally categorized 252 SNVs from exon 13 into 188 functional and 60 non-functional variants, demonstrating that saturation genome editing (SGE) coupled with drug sensitivity assays can enhance functional annotation of BRCA2 VUS.

Gene-specific machine learning for pathogenicity prediction of rare BRCA1 and BRCA2 missense variants

Machine learning-based pathogenicity prediction helps interpret rare missense variants of BRCA1 and BRCA2, which are associated with hereditary cancers. Recent studies have shown that classifiers trained using variants of a specific gene or a set of genes related to a particular disease perform better than those trained using all variants, due to their higher specificity, despite the smaller training dataset size. In this study, we further investigated the advantages of "gene-specific" machine learning compared to "disease-specific" machine learning. We used 1068 rare (gnomAD minor allele frequency (MAF) < 0.005) missense variants of 28 genes associated with hereditary cancers for our investigation. Popular machine learning classifiers were employed: regularized logistic regression, extreme gradient boosting, random forests, support vector machines, and deep neural networks. As features, we used MAFs from multiple populations, functional prediction and conservation scores, and positions of variants. The disease-specific training dataset included the gene-specific training dataset and was > 7 × larger. However, we observed that gene-specific training variants were sufficient to produce the optimal pathogenicity predictor if a suitable machine learning classifier was employed. Therefore, we recommend gene-specific over disease-specific machine learning as an efficient and effective method for predicting the pathogenicity of rare BRCA1 and BRCA2 missense variants.

Functional genomics of OCTN2 variants informs protein-specific variant effect predictor for Carnitine Transporter Deficiency

Genetic variants in SLC22A5, encoding the membrane carnitine transporter OCTN2, cause the rare metabolic disorder Carnitine Transporter Deficiency (CTD). CTD is potentially lethal but actionable if detected early, with confirmatory diagnosis involving sequencing of SLC22A5. Interpretation of missense variants of uncertain significance (VUSs) is a major challenge. In this study, we sought to characterize the largest set to date (n = 150) of OCTN2 variants identified in diverse ancestral populations, with the goals of furthering our understanding of the mechanisms leading to OCTN2 loss-of-function (LOF) and creating a protein-specific variant effect prediction model for OCTN2 function. Uptake assays with 14C-carnitine revealed that 105 variants (70%) significantly reduced transport of carnitine compared to wild-type OCTN2, and 37 variants (25%) severely reduced function to less than 20%. All ancestral populations harbored LOF variants; 62% of green fluorescent protein (GFP)-tagged variants impaired OCTN2 localization to the plasma membrane of human embryonic kidney (HEK293T) cells, and subcellular localization significantly associated with function, revealing a major LOF mechanism of interest for CTD. With these data, we trained a model to classify variants as functional (>20% function) or LOF (<20% function). Our model outperformed existing state-of-the-art methods as evaluated by multiple performance metrics, with mean area under the receiver operating characteristic curve (AUROC) of 0.895 ± 0.025. In summary, in this study we generated a rich dataset of OCTN2 variant function and localization, revealed important disease-causing mechanisms, and improved upon machine learning-based prediction of OCTN2 variant function to aid in variant interpretation in the diagnosis and treatment of CTD.

Variant predictions in congenital adrenal hyperplasia caused by mutations in CYP21A2

CYP21A2 deficiency represents 95% of congenital adrenal hyperplasia (CAH) cases, a group of genetic disorders that affect steroid biosynthesis. The genetic and functional analysis provide critical tools to elucidate complex CAH cases. One of the most accessible tools to infer the pathogenicity of new variants is in silico prediction. Here, we analyzed the performance of in silico prediction tools to categorize missense single nucleotide variants (SNVs) of CYP21A2. SNVs of CYP21A2 characterized in vitro by functional assays were selected to assess the performance of online single and meta predictors. SNVs were tested separately or in combination with the related phenotype (severe or mild CAH form). In total, 103 SNVs of CYP21A2 (90 pathogenic and 13 neutral) were used to test the performance of 13 single-predictors and four meta-predictors. All SNVs associated with the severe phenotypes were well categorized by all tools, with an accuracy of between 0.69 (PredictSNP2) and 0.97 (CADD), and Matthews’ correlation coefficient (MCC) between 0.49 (PoredicSNP2) and 0.90 (CADD). However, SNVs related to the mild phenotype had more variation, with the accuracy between 0.47 (S3Ds&amp;GO and MAPP) and 0.88 (CADD), and MCC between 0.18 (MAPP) and 0.71 (CADD). From our analysis, we identified four predictors of CYP21A2 variant pathogenicity with good performance, CADD, ConSurf, DANN, and PolyPhen2. These results can be used for future analysis to infer the impact of uncharacterized SNVs in CYP21A2.

Gene-specific machine learning model to predict the pathogenicity of BRCA2 variants

Background: Existing BRCA2-specific variant pathogenicity prediction algorithms focus on the prediction of the functional impact of a subtype of variants alone. General variant effect predictors are applicable to all subtypes, but are trained on putative benign and pathogenic variants and do not account for gene-specific information, such as hotspots of pathogenic variants. Local, gene-specific information have been shown to aid variant pathogenicity prediction; therefore, our aim was to develop a BRCA2-specific machine learning model to predict pathogenicity of all types of BRCA2 variants.

Methods: We developed an XGBoost-based machine learning model to predict pathogenicity of BRCA2 variants. The model utilizes general variant information such as position, frequency, and consequence for the canonical BRCA2 transcript, as well as deleteriousness prediction scores from several tools. We trained the model on 80% of the expert reviewed variants by the Evidence-Based Network for the Interpretation of Germline Mutant Alleles (ENIGMA) consortium and tested its performance on the remaining 20%, as well as on an independent set of variants of uncertain significance with experimentally determined functional scores.

Results: The novel gene-specific model predicted the pathogenicity of ENIGMA BRCA2 variants with an accuracy of 99.9%. The model also performed excellently on predicting the functional consequence of the independent set of variants (accuracy was up to 91.3%).

Conclusion: This new, gene-specific model is an accurate method for interpreting the pathogenicity of variants in the BRCA2 gene. It is a valuable addition for variant classification and can prioritize unreviewed variants for functional analysis or expert review.

Classification of BRCA2 Variants of Uncertain Significance (VUS) Using an ACMG/AMP Model Incorporating a Homology-Directed Repair (HDR) Functional Assay

PURPOSE

The identification of variants of uncertain significance (VUS) in the BRCA1 and BRCA2 genes by hereditary cancer testing poses great challenges for the clinical management of variant carriers. The ACMG/AMP (American College of Medical Genetics and Genomics/Association for Molecular Pathology) variant classification framework, which incorporates multiple sources of evidence, has the potential to establish the clinical relevance of many VUS. We sought to classify the clinical relevance of 133 single-nucleotide substitution variants encoding missense variants in the DNA-binding domain (DBD) of BRCA2 by incorporating results from a validated functional assay into an ACMG/AMP-variant classification model from a hereditary cancer-testing laboratory.

EXPERIMENTAL DESIGN

The 133 selected VUS were evaluated using a validated homology-directed double-strand DNA break repair (HDR) functional assay. Results were combined with clinical and genetic data from variant carriers in a rules-based variant classification model for BRCA2.

RESULTS

Of 133 missense variants, 44 were designated as non-functional and 89 were designated as functional in the HDR assay. When combined with genetic and clinical information from a single diagnostic laboratory in an ACMG/AMP-variant classification framework, 66 variants previously classified by the diagnostic laboratory were correctly classified, and 62 of 67 VUS (92.5%) were reclassified as likely pathogenic (n = 22) or likely benign (n = 40). In total, 44 variants were classified as pathogenic/likely pathogenic, 84 as benign/likely benign, and 5 remained as VUS.

CONCLUSIONS

Incorporation of HDR functional analysis into an ACMG/AMP framework model substantially improves BRCA2 VUS re-classification and provides an important tool for determining the clinical relevance of individual BRCA2 VUS.

Understanding and predicting the functional consequences of missense mutations in BRCA1 and BRCA2

BRCA1 and BRCA2 are tumour suppressor genes that play a critical role in maintaining genomic stability via the DNA repair mechanism. DNA repair defects caused by BRCA1 and BRCA2 missense variants increase the risk of developing breast and ovarian cancers. Accurate identification of these variants becomes clinically relevant, as means to guide personalized patient management and early detection. Next-generation sequencing efforts have significantly increased data availability but also the discovery of variants of uncertain significance that need interpretation. Experimental approaches used to measure the molecular consequences of these variants, however, are usually costly and time-consuming. Therefore, computational tools have emerged as faster alternatives for assisting in the interpretation of the clinical significance of newly discovered variants. To better understand and predict variant pathogenicity in BRCA1 and BRCA2, various machine learning algorithms have been proposed, however presented limited performance. Here we present BRCA1 and BRCA2 gene-specific models and a generic model for quantifying the functional impacts of single-point missense variants in these genes. Across tenfold cross-validation, our final models achieved a Matthew's Correlation Coefficient (MCC) of up to 0.98 and comparable performance of up to 0.89 across independent, non-redundant blind tests, outperforming alternative approaches. We believe our predictive tool will be a valuable resource for providing insights into understanding and interpreting the functional consequences of missense variants in these genes and as a tool for guiding the interpretation of newly discovered variants and prioritizing mutations for experimental validation.

Validation and Data-Integration of Yeast-Based Assays for Functional Classification of BRCA1 Missense Variants

Germline mutations in the BRCA1 gene have been reported to increase the lifetime risk of developing breast and/or ovarian cancer (BOC). By new sequencing technologies, numerous variants of uncertain significance (VUS) are identified. It is mandatory to develop new tools to evaluate their functional impact and pathogenicity. As the expression of pathogenic BRCA1 variants in Saccharomyces cerevisiae increases the frequency of intra- and inter-chromosomal homologous recombination (HR), and gene reversion (GR), we validated the two HR and the GR assays by testing 23 benign and 23 pathogenic variants and compared the results with those that were obtained in the small colony phenotype (SCP) assay, an additional yeast-based assay, that was validated previously. We demonstrated that they scored high accuracy, sensitivity, and sensibility. By using a classifier that was based on majority of voting, we have integrated data from HR, GR, and SCP assays and developed a reliable method, named yBRCA1, with high sensitivity to obtain an accurate VUS functional classification (benign or pathogenic). The classification of BRCA1 variants, important for assessing the risk of developing BOC, is often difficult to establish with genetic methods because they occur rarely in the population. This study provides a new tool to get insights on the functional impact of the BRCA1 variants.

Use of machine learning to classify high-risk variants of uncertain significance in lamin A/C cardiac disease

BACKGROUND

Variation in lamin A/C results in a spectrum of clinical disease, including arrhythmias and cardiomyopathy. Benign variation is rare, and classification of LMNA missense variants via in silico prediction tools results in a high rate of variants of uncertain significance (VUSs).

OBJECTIVE

The goal of this study was to use a machine learning (ML) approach for in silico prediction of LMNA pathogenic variation.

METHODS

Genetic sequencing was performed on family members with conduction system disease, and patient cell lines were examined for LMNA expression. In silico predictions of conservation and pathogenicity of published LMNA variants were visualized with uniform manifold approximation and projection. K-means clustering was used to identify variant groups with similarly projected scores, allowing the generation of statistically supported risk categories.

RESULTS

We discovered a novel LMNA variant (c.408C>A:p.Asp136Glu) segregating with conduction system disease in a multigeneration pedigree, which was reported as a VUS by a commercial testing company. Additional familial analysis and in vitro testing found it to be pathogenic, which prompted the development of an ML algorithm that used in silico predictions of pathogenicity for known LMNA missense variants. This identified 3 clusters of variation, each with a significantly different incidence of known pathogenic variants (38.8%, 15.0%, and 6.1%). Three hundred thirty-nine of 415 head/rod domain variants (81.7%), including p.Asp136Glu, were in clusters with highest proportions of pathogenic variants.

CONCLUSION

An unsupervised ML method successfully identified clusters enriched for pathogenic LMNA variants including a novel variant associated with conduction system disease. Our ML method may assist in identifying high-risk VUS when familial testing is unavailable.

Molecular-based precision oncology clinical decision making augmented by artificial intelligence

The rapid growth and decreasing cost of Next-generation sequencing (NGS) technologies have made it possible to conduct routine large panel genomic sequencing in many disease settings, especially in the oncology domain. Furthermore, it is now known that optimal disease management of patients depends on individualized cancer treatment guided by comprehensive molecular testing. However, translating results from molecular sequencing reports into actionable clinical insights remains a challenge to most clinicians. In this review, we discuss about some representative systems that leverage artificial intelligence (AI) to facilitate some processes of clinicians' decision making based upon molecular data, focusing on their application in precision oncology. Some limitations and pitfalls of the current application of AI in clinical decision making are also discussed.

DNA double-strand break repair in cancer: A path to achieving precision medicine

The assessment of DNA damage can be a significant diagnostic for precision medicine. DNA double strand break (DSBs) pathways in cancer are the primary targets in a majority of anticancer therapies, yet the molecular vulnerabilities that underlie each tumor can vary widely making the application of precision medicine challenging. Identifying and understanding these interindividual vulnerabilities enables the design of targeted DSB inhibitors along with evolving precision medicine approaches to selectively kill cancer cells with minimal side effects. A major challenge however, is defining exactly how to target unique differences in DSB repair pathway mechanisms. This review comprises a brief overview of the DSB repair mechanisms in cancer and includes results obtained with revolutionary advances such as CRISPR/Cas9 and machine learning/artificial intelligence, which are rapidly advancing not only our understanding of determinants of DSB repair choice, but also how it can be used to advance precision medicine. Scientific innovation in the methods used to diagnose and treat cancer is converging with advances in basic science and translational research. This revolution will continue to be a critical driver of precision medicine that will enable precise targeting of unique individual mechanisms. This review aims to lay the foundation for achieving this goal.

Computational Tools to Assess the Functional Consequences of Rare and Noncoding Pharmacogenetic Variability

Interindividual differences in drug response are a common concern in both drug development and across layers of care. While genetics clearly influences drug response and toxicity of many drugs, a substantial fraction of the heritable pharmacological and toxicological variability remains unexplained by known genetic polymorphisms. In recent years, population-scale sequencing projects have unveiled tens of thousands of coding and noncoding pharmacogenetic variants with unclear functional effects that might explain at least part of this missing heritability. However, translating these personalized variant signatures into drug response predictions and actionable advice remains challenging and constitutes one of the most important frontiers of contemporary pharmacogenomics. Conventional prediction methods are primarily based on evolutionary conservation, which drastically reduces their predictive accuracy when applied to poorly conserved pharmacogenes. Here, we review the current state-of-the-art of computational variant effect predictors across variant classes and critically discuss their utility for pharmacogenomics. Besides missense variants, we discuss recent progress in the evaluation of synonymous, splice, and noncoding variations. Furthermore, we discuss emerging possibilities to assess haplotypes and structural variations. We advocate for the development of algorithms trained on pharmacogenomic instead of pathogenic data sets to improve the predictive accuracy in order to facilitate the utilization of next-generation sequencing data for personalized clinical decision support and precision pharmacogenomics.

Strong functional data for pathogenicity or neutrality classify BRCA2 DNA-binding-domain variants of uncertain significance

Determination of the clinical relevance of rare germline variants of uncertain significance (VUSs) in the BRCA2 cancer predisposition gene remains a challenge as a result of limited availability of data for use in classification models. However, laboratory-based functional data derived from validated functional assays of known sensitivity and specificity may influence the interpretation of VUSs. We evaluated 252 missense VUSs from the BRCA2 DNA-binding domain by using a homology-directed DNA repair (HDR) assay and identified 90 as non-functional and 162 as functional. The functional assay results were integrated with other available data sources into an ACMG/AMP rules-based classification framework used by a hereditary cancer testing laboratory. Of the 186 missense variants observed by the testing laboratory, 154 were classified as VUSs without functional data. However, after applying protein functional data, 86% (132/154) of the VUSs were reclassified as either likely pathogenic/pathogenic (39/132) or likely benign/benign (93/132), which impacted testing results for 1,900 individuals. These results indicate that validated functional assay data can have a substantial impact on VUS classification and associated clinical management for many individuals with inherited alterations in BRCA2.

Lessons learned from understanding chemotherapy resistance in epithelial tubo-ovarian carcinoma from BRCA1and BRCA2mutation carriers

BRCA1 and BRCA2 are multi-functional proteins and key factors for maintaining genomic stability through their roles in DNA double strand break repair by homologous recombination, rescuing stalled or damaged DNA replication forks, and regulation of cell cycle DNA damage checkpoints. Impairment of any of these critical roles results in genomic instability, a phenotypic hallmark of many cancers including breast and epithelial ovarian carcinomas (EOC). Damaging, usually loss of function germline and somatic variants in BRCA1 and BRCA2, are important drivers of the development, progression, and management of high-grade serous tubo-ovarian carcinoma (HGSOC). However, mutations in these genes render patients particularly sensitive to platinum-based chemotherapy, and to the more innovative targeted therapies with poly-(ADP-ribose) polymerase inhibitors (PARPis) that are targeted to BRCA1/BRCA2 mutation carriers. Here, we reviewed the literature on the responsiveness of BRCA1/2-associated HGSOC to platinum-based chemotherapy and PARPis, and propose mechanisms underlying the frequent development of resistance to these therapeutic agents.