Evaluating the impact of in silico predictors on clinical variant classification

Explicable prioritization of genetic variants by integration of rule-based and machine learning algorithms for diagnosis of rare Mendelian disorders

BACKGROUND

In the process of finding the causative variant of rare diseases, accurate assessment and prioritization of genetic variants is essential. Previous variant prioritization tools mainly depend on the in-silico prediction of the pathogenicity of variants, which results in low sensitivity and difficulty in interpreting the prioritization result. In this study, we propose an explainable algorithm for variant prioritization, named 3ASC, with higher sensitivity and ability to annotate evidence used for prioritization. 3ASC annotates each variant with the 28 criteria defined by the ACMG/AMP genome interpretation guidelines and features related to the clinical interpretation of the variants. The system can explain the result based on annotated evidence and feature contributions.

RESULTS

We trained various machine learning algorithms using in-house patient data. The performance of variant ranking was assessed using the recall rate of identifying causative variants in the top-ranked variants. The best practice model was a random forest classifier that showed top 1 recall of 85.6% and top 3 recall of 94.4%. The 3ASC annotates the ACMG/AMP criteria for each genetic variant of a patient so that clinical geneticists can interpret the result as in the CAGI6 SickKids challenge. In the challenge, 3ASC identified causal genes for 10 out of 14 patient cases, with evidence of decreased gene expression for 6 cases. Among them, two genes (HDAC8 and CASK) had decreased gene expression profiles confirmed by transcriptome data.

CONCLUSIONS

3ASC can prioritize genetic variants with higher sensitivity compared to previous methods by integrating various features related to clinical interpretation, including features related to false positive risk such as quality control and disease inheritance pattern. The system allows interpretation of each variant based on the ACMG/AMP criteria and feature contribution assessed using explainable AI techniques.

Evaluating ClinGen variant curation expert panels' application of PVS1 code

BACKGROUND

The 2015 American College of Medical Genetics and Genomics (ACMG) and Association for Molecular Pathology (AMP) guidelines articulates that the effects of certain types of variants on gene function can often be seen as a complete absence of the gene product by leading to a lack of transcription or nonsense-mediated decay(NMD). However, detailed information considering different types of loss of function(LOF) variants, refined steps assimilating details concerning location of variant, changes in strength levels, NMD boundary, or any additional information pointing to a true null effect, were all left to expert judgement. As part of its Clinical Genome Resource (ClinGen) initiative, Variant Curation Expert Panels (VCEPs) are designated to make gene/disease-centric specifications in accordance with the ACMG/AMP guidelines, including a more detailed definition of what constitutes an appropriate LOF evidence. Our goal was to evaluate the current LOF guidelines developed by the VCEPs and analyse the prior curated variants concerning the PVS1 criteria, bringing people occupied in genetic data analysis a comprehensive understanding of this code.

METHODS

Our study evaluated 7 VCEPs for their LOF criteria (PVS1). Subsequently, we assessed the predictive criteria by considering the underlying disease mechanism, protein transcript, and variant types delineated. Then, we meticulously curated the LOF evidence referenced by each VCEP in their preliminary variant classification, thereby scrutinizing the recommendations put forth by VCEPs and their application in the interpretation of the aforementioned predictive criteria. Based on these, an extensive curation of evidence summary considering PVS1 applied by VCEPs according to their classification of pilot variants for the purpose of analyzing VCEP criteria specifications and their use in the understanding of LOF was conducted.

RESULTS

We observed in this article that the VCEPs discussed followed the majority of Sequence Variant Interpretation (SVI) recommendations concerning the application of this LOF criteria, except for some disease/gene specific considerations. We highlighted the wide range of PVS1 strength levels approved by VCEP, reflecting the diversity of evidence for each variants type. In addition, we observed substantial differences in the approach used to determine relative strengths for different types of null variants and in the attitude towards these principles concerning variant location, NMD and influence to protein function between VCEPs.

CONCLUSIONS

It is difficult to understand the intricacies of the predictive data(PVS1), which often requires expert-level knowledge of disease/gene. The VCEP criteria specifications for the predictive evidence play an important role in making it more accessible for the curators to apply the predictive data by providing details concerning this complex criteria. Despite this, we believe there is a need for more guidance on standardizing this process and ensuring consistency in the application of this predictive evidence.

Next-generation sequencing and bioinformatics in rare movement disorders

The ability to sequence entire exomes and genomes has revolutionized molecular testing in rare movement disorders, and genomic sequencing is becoming an integral part of routine diagnostic workflows for these heterogeneous conditions. However, interpretation of the extensive genomic variant information that is being generated presents substantial challenges. In this Perspective, we outline multidimensional strategies for genetic diagnosis in patients with rare movement disorders. We examine bioinformatics tools and computational metrics that have been developed to facilitate accurate prioritization of disease-causing variants. Additionally, we highlight community-driven data-sharing and case-matchmaking platforms, which are designed to foster the discovery of new genotype-phenotype relationships. Finally, we consider how multiomic data integration might optimize diagnostic success by combining genomic, epigenetic, transcriptomic and/or proteomic profiling to enable a more holistic evaluation of variant effects. Together, the approaches that we discuss offer pathways to the improved understanding of the genetic basis of rare movement disorders.

MAGPIE: accurate pathogenic prediction for multiple variant types using machine learning approach

Identifying pathogenic variants from the vast majority of nucleotide variation remains a challenge. We present a method named Multimodal Annotation Generated Pathogenic Impact Evaluator (MAGPIE) that predicts the pathogenicity of multi-type variants. MAGPIE uses the ClinVar dataset for training and demonstrates superior performance in both the independent test set and multiple orthogonal validation datasets, accurately predicting variant pathogenicity. Notably, MAGPIE performs best in predicting the pathogenicity of rare variants and highly imbalanced datasets. Overall, results underline the robustness of MAGPIE as a valuable tool for predicting pathogenicity in various types of human genome variations. MAGPIE is available at https://github.com/shenlab-genomics/magpie .

Functional characterization of all CDKN2A missense variants and comparison to in silico models of pathogenicity

Interpretation of variants identified during genetic testing is a significant clinical challenge. In this study, we developed a high-throughput CDKN2A functional assay and characterized all possible CDKN2A missense variants. We found that 40% of all missense variants were functionally deleterious. We also used our functional classification to assess the performance of in silico models that predict the effect of variants, including recently reported models based on machine learning. Notably, we found that all in silico models similarly when compared to our functional classifications with accuracies of 54.6 - 70.9%. Furthermore, while we found that functionally deleterious variants were enriched within ankyrin repeats, rarely were all missense variants at a single residue functionally deleterious. Our functional classifications are a resource to aid the interpretation of CDKN2A variants and have important implications for the application of variant interpretation guidelines, particularly the use of in silico models for clinical variant interpretation.

Compilation of Genotype and Phenotype Data in GCDH-LOVD for Variant Classification and Further Application

Glutaric aciduria type 1 (GA-1) is a rare but treatable autosomal-recessive neurometabolic disorder of lysin metabolism caused by biallelic pathogenic variants in glutaryl-CoA dehydrogenase gene (GCDH) that lead to deficiency of GCDH protein. Without treatment, this enzyme defect causes a neurological phenotype characterized by movement disorder and cognitive impairment. Based on a comprehensive literature search, we established a large dataset of GCDH variants using the Leiden Open Variation Database (LOVD) to summarize the known genotypes and the clinical and biochemical phenotypes associated with GA-1. With these data, we developed a GCDH-specific variation classification framework based on American College of Medical Genetics and Genomics and the Association for Molecular Pathology guidelines. We used this framework to reclassify published variants and to describe their geographic distribution, both of which have practical implications for the molecular genetic diagnosis of GA-1. The freely available GCDH-specific LOVD dataset provides a basis for diagnostic laboratories and researchers to further optimize their knowledge and molecular diagnosis of this rare disease.

Finding significance: New perspectives in variant classification of the RAD51 regulators, BRCA2 and beyond

For many individuals harboring a variant of uncertain functional significance (VUS) in a homologous recombination (HR) gene, their risk of developing breast and ovarian cancer is unknown. Integral to the process of HR are BRCA1 and regulators of the central HR protein, RAD51, including BRCA2, PALB2, RAD51C and RAD51D. Due to advancements in sequencing technology and the continued expansion of cancer screening panels, the number of VUS identified in these genes has risen significantly. Standard practices for variant classification utilize different types of predictive, population, phenotypic, allelic and functional evidence. While variant analysis is improving, there remains a struggle to keep up with demand. Understanding the effects of an HR variant can aid in preventative care and is critical for developing an effective cancer treatment plan. In this review, we discuss current perspectives in the classification of variants in the breast and ovarian cancer genes BRCA1, BRCA2, PALB2, RAD51C and RAD51D.

O’Brien T, Perucca P, Ge Z, Anderson A. Applications for Deep Learning in Epilepsy Genetic Research

Epilepsy is a group of brain disorders characterised by an enduring predisposition to generate unprovoked seizures. Fuelled by advances in sequencing technologies and computational approaches, more than 900 genes have now been implicated in epilepsy. The development and optimisation of tools and methods for analysing the vast quantity of genomic data is a rapidly evolving area of research. Deep learning (DL) is a subset of machine learning (ML) that brings opportunity for novel investigative strategies that can be harnessed to gain new insights into the genomic risk of people with epilepsy. DL is being harnessed to address limitations in accuracy of long-read sequencing technologies, which improve on short-read methods. Tools that predict the functional consequence of genetic variation can represent breaking ground in addressing critical knowledge gaps, while methods that integrate independent but complimentary data enhance the predictive power of genetic data. We provide an overview of these DL tools and discuss how they may be applied to the analysis of genetic data for epilepsy research.

High diagnostic yield of targeted next-generation sequencing panel as a first-tier molecular test for the patients with myopathy or muscular dystrophy

Muscular dystrophies are a heterogeneous group of neuromuscular disorders with a wide range of the clinical and genetic spectrum. Whole-exome sequencing (WES) has been on the rise to become the usual method of choice for molecular diagnosis in patients presenting with muscular dystrophy or congenital or metabolic myopathy phenotype. Here, we used a panel with 47 genes including not only muscular dystrophy but also myopathy-associated genes that had been used as a first-tier approach. A total of 146 patients who were referred to our clinic with the prediagnosis of muscular dystrophy and/or myopathy were included in the study. Dystrophin gene deletion/duplication was ruled out on the patients with a preliminary diagnosis of Duchenne muscular dystrophy. In this study, the molecular etiology of 67 patients was proved with the gene panel with a diagnostic yield of 46%. Causal variants were identified in 23 genes including CAPN3(11), DYSF(9), DMD(8), SGCA(5), TTN(4), LAMA2(3), LMNA(3), SGCB(3), COL6A1(3), DES (2), CAV3(2), FKRP(2), FKTN(2), ANO5, COL6A2, CLCN1, GNE, POMGNT1, POMGNT2, POMT2, SYNE1, TCAP, and FLNC with 16 novel variants. There were 27 patients with uncertain molecular results including the ones who had a variant of uncertain significance, who had only one heterozygous variant for an autosomal recessive disease, and the ones who had two variants in different genes. Molecular diagnosis in muscular dystrophy is essential to plan clinical management and choosing treatment options. Also, the results will affect the reproduction options. Targeted next-generation sequencing is a cost-effective method that reduces the WES requirements with a significant diagnostic rate.

Multiple endocrine neoplasia type 2 (MEN2) and RET specific modifications of the ACMG/AMP variant classification guidelines and impact on the MEN2 RET database

The Multiple Endocrine Neoplasia type 2 (MEN2) RET proto-oncogene database, originally published in 2008, is a comprehensive repository of all publicly available RET gene variations associated with MEN2 syndromes. The variant-specific genotype/phenotype information, age of earliest reported medullary thyroid carcinoma (MTC) onset, and relevant references with a brief summary of findings are cataloged. The ACMG/AMP 2015 consensus statement on variant classification was modified specifically for MEN2 syndromes and RET variants using ClinGen sequence variant interpretation working group recommendations and ClinGen expert panel manuscripts, as well as manuscripts from the American Thyroid Association Guidelines Task Force on Medullary Thyroid Carcinoma and other MEN2 RET literature. The classifications for the 166 single unique variants in the MEN2 RET database were reanalyzed using the MEN2 RET specifically modified ACMG/AMP classification guidelines (version 1). Applying these guidelines added two new variant classifications to the database (likely benign and likely pathogenic) and resulted in clinically significant classification changes (e.g., from pathogenic to uncertain) in 15.7% (26/166) of the original variants. Of those clinically significant changes, the highest percentage of changes, 46.2% (12/26), were changes from uncertain to benign or likely benign. The modified ACMG/AMP criteria with MEN2 RET specifications will optimize and standardize RET variant classifications.

Insights on variant analysis in silico tools for pathogenicity prediction

Molecular biology is currently a fast-advancing science. Sequencing techniques are getting cheaper, but the interpretation of genetic variants requires expertise and computational power, therefore is still a challenge. Next-generation sequencing releases thousands of variants and to classify them, researchers propose protocols with several parameters. Here we present a review of several in silico pathogenicity prediction tools involved in the variant prioritization/classification process used by some international protocols for variant analysis and studies evaluating their efficiency.

Phenotypic screening models for rapid diagnosis of genetic variants and discovery of personalized therapeutics

Precision medicine strives for highly individualized treatments for disease under the notion that each individual's unique genetic makeup and environmental exposures imprints upon them not only a disposition to illness, but also an optimal therapeutic approach. In the realm of rare disorders, genetic predisposition is often the predominant mechanism driving disease presentation. For such, mostly, monogenic disorders, a causal gene to phenotype association is likely. As a result, it becomes important to query the patient's genome for the presence of pathogenic variations that are likely to cause the disease. Determining whether a variant is pathogenic or not is critical to these analyses and can be challenging, as many disease-causing variants are novel and, ergo, have no available functional data to help categorize them. This problem is exacerbated by the need for rapid evaluation of pathogenicity, since many genetic diseases present in young children who will experience increased morbidity and mortality without rapid diagnosis and therapeutics. Here, we discuss the utility of animal models, with a focus mainly on C. elegans, as a contrast to tissue culture and in silico approaches, with emphasis on how these systems are used in determining pathogenicity of variants with uncertain significance and then used to screen for novel therapeutics.

Splicing Analysis of 16 PALB2 ClinVar Variants by Minigene Assays: Identification of Six Likely Pathogenic Variants

PALB2 loss-of-function variants are associated with significant increased risk of breast cancer as well as other types of tumors. Likewise, splicing disruptions are a common mechanism of disease susceptibility. Indeed, we previously showed, by minigene assays, that 35 out of 42 PALB2 variants impaired splicing. Taking advantage of one of these constructs (mgPALB2_ex1-3), we proceeded to analyze other variants at exons 1 to 3 reported at the ClinVar database. Thirty-one variants were bioinformatically analyzed with MaxEntScan and SpliceAI. Then, 16 variants were selected for subsequent RNA assays. We identified a total of 12 spliceogenic variants, 11 of which did not produce any trace of the expected minigene full-length transcript. Interestingly, variant c.49-1G > A mimicked previous outcomes in patient RNA (transcript ∆(E2p6)), supporting the reproducibility of the minigene approach. A total of eight variant-induced transcripts were characterized, three of which (∆(E1q17), ∆(E3p11), and ∆(E3)) were predicted to introduce a premature termination codon and to undergo nonsense-mediated decay, and five (▼(E1q9), ∆(E2p6), ∆(E2), ▼(E3q48)-a, and ▼(E3q48)-b) maintained the reading frame. According to an ACMG/AMP (American College of Medical Genetics and Genomics/Association for Molecular Pathology)-based classification scheme, which integrates mgPALB2 data, six PALB2 variants were classified as pathogenic/likely pathogenic, five as VUS, and five as likely benign. Furthermore, five ±1,2 variants were catalogued as VUS because they produced significant proportions of in-frame transcripts of unknown impact on protein function.

A massively parallel assay accurately discriminates between functionally normal and abnormal variants in a hotspot domain of KCNH2

Many genes, including KCNH2, contain "hotspot" domains associated with a high density of variants associated with disease. This has led to the suggestion that variant location can be used as evidence supporting classification of clinical variants. However, it is not known what proportion of all potential variants in hotspot domains cause loss of function. Here, we have used a massively parallel trafficking assay to characterize all single-nucleotide variants in exon 2 of KCNH2, a known hotspot for variants that cause long QT syndrome type 2 and an increased risk of sudden cardiac death. Forty-two percent of KCNH2 exon 2 variants caused at least 50% reduction in protein trafficking, and 65% of these trafficking-defective variants exerted a dominant-negative effect when co-expressed with a WT KCNH2 allele as assessed using a calibrated patch-clamp electrophysiology assay. The massively parallel trafficking assay was more accurate (AUC of 0.94) than bioinformatic prediction tools (REVEL and CardioBoost, AUC of 0.81) in discriminating between functionally normal and abnormal variants. Interestingly, over half of variants in exon 2 were found to be functionally normal, suggesting a nuanced interpretation of variants in this "hotspot" domain is necessary. Our massively parallel trafficking assay can provide this information prospectively.

Association of RNF213 Variants With Periventricular Anastomosis in Moyamoya Disease

BACKGROUND

The pathogenic mechanisms of periventricular anastomosis (PA) in moyamoya disease remain unknown. Here, we aimed to describe the angiographic profiles of PA and their relationships with really interesting new gene (RING) finger protein 213 (RNF213) genotypes.

METHODS

We conducted a retrospective cohort study of moyamoya disease patients consecutively recruited between June 2019 and January 2021 in Beijing Tiantan Hospital, Capital Medical University, China. C-terminal region of RNF213 was sequenced. Angiographic characteristics of PA vessels (lenticulostriate artery, thalamotuberal artery, thalamoperforating artery, anterior choroidal artery, and posterior choroidal artery) were compared between different groups of RNF213 genotypes. The dilatation and extension of PA vessels were measured by using PA score (positive, score 1-5; negative, score 0). Multivariate regression analysis was conducted to assess variables associated with PA score. In addition, gene expression of RNF213 in human brain regions was evaluated from the Allen Human Brain Atlas.

RESULTS

Among 260 patients (484 hemispheres), 71.2% carried no RNF213 rare and novel variants, 20.0% carried p.R4810K heterozygotes, and 8.8% carried other rare and novel variants. PA scores in patients with p.R4810K and other rare and novel variants were significantly higher than in wild-type patients (P<0.001). Age (odds ratio [OR], 0.958 [95% CI, 0.942-0.974]; P<0.001), platelet count (OR, 0.996 [95% CI, 0.992-0.999]; P=0.027), p.R4810K variant (OR, 2.653 [95% CI, 1.514-4.649]; P=0.001), other rare and novel variants (OR, 3.197 [95% CI, 1.012-10.094]; P=0.048), Suzuki stage ≥4 (OR, 1.941 [95% CI, 1.138-3.309]; P=0.015), and posterior cerebral artery involvement (OR, 1.827 [95% CI, 1.020-3.271]; P=0.043) were significantly correlated with PA score. High expression of RNF213 was detected in the periventricular area.

CONCLUSIONS

RNF213 variants were confirmed to be associated with PA in moyamoya disease. Individuals with RNF213 p.R4810K heterozygotes and other C-terminal region rare variants exhibited different angiographic phenotypes, compared with wild-type patients.

Re-evaluation of Missense Variant Classifications in NF2

Missense variants in the NF2 gene result in variable NF2 disease presentation. Clinical classification of missense variants often represents a challenge, due to lack of evidence for pathogenicity and function. This study provides a summary of NF2 missense variants, with variant classifications based on currently available evidence. NF2 missense variants were collated from pathology‐associated databases and existing literature. Association for Clinical Genomic Sciences Best Practice Guidelines (2020) were followed in the application of evidence for variant interpretation and classification. The majority of NF2 missense variants remain classified as variants of uncertain significance. However, NF2 missense variants identified in gnomAD occurred at a consistent rate across the gene, while variants compiled from pathology‐associated databases displayed differing rates of variation by exon of NF2. The highest rate of NF2 disease‐associated variants was observed in exon 7, while lower rates were observed toward the C‐terminus of the NF2 protein, merlin. Further phenotypic information associated with variants, alongside variant‐specific functional analysis, is necessary for more definitive variant interpretation. Our data identified differences in frequency of NF2 missense variants by exon between gnomAD population data and NF2 disease‐associated variants, suggesting a potential genotype‐phenotype correlation; further work is necessary to substantiate this.

Evaluation of in silico predictors on short nucleotide variants in HBA1, HBA2, and HBB associated with haemoglobinopathies

Haemoglobinopathies are the commonest monogenic diseases worldwide and are caused by variants in the globin gene clusters. With over 2400 variants detected to date, their interpretation using the American College of Medical Genetics and Genomics (ACMG)/Association for Molecular Pathology (AMP) guidelines is challenging and computational evidence can provide valuable input about their functional annotation. While many in silico predictors have already been developed, their performance varies for different genes and diseases. In this study, we evaluate 31 in silico predictors using a dataset of 1627 variants in HBA1, HBA2, and HBB. By varying the decision threshold for each tool, we analyse their performance (a) as binary classifiers of pathogenicity and (b) by using different non-overlapping pathogenic and benign thresholds for their optimal use in the ACMG/AMP framework. Our results show that CADD, Eigen-PC, and REVEL are the overall top performers, with the former reaching moderate strength level for pathogenic prediction. Eigen-PC and REVEL achieve the highest accuracies for missense variants, while CADD is also a reliable predictor of non-missense variants. Moreover, SpliceAI is the top performing splicing predictor, reaching strong level of evidence, while GERP++ and phyloP are the most accurate conservation tools. This study provides evidence about the optimal use of computational tools in globin gene clusters under the ACMG/AMP framework.