Human Mutation

Centre for Genomic Regulation (CRG), The Barcelona Institute of Science and Technology, Barcelona, Spain Universitat Pompeu Fabra (UPF), Barcelona, Spain Barcelona Supercomputing Center (BSC), Barcelona, Spain Bioinformatics Unit (MF1), Department for Methods Development and Research Infrastructure, Robert Koch Institute, Berlin, Germany NKI Netherlands Cancer Institute, The Netherlands Institut de Recerca Sant Joan de Déu, University of Barcelona, Barcelona, Spain Department of Neurology, University Hospital Zurich, Zurich, Switzerland L'Institut du Thorax, INSERM, CNRS, Univ Nantes, Nantes, France Service de Cardiologie, L'institut du thorax, CHU Nantes, Nantes, France Vall d'Hebron Institut de Recerca (VHIR), Barcelona, Spain Sidra Medicine, Doha, Qatar Women's Health Dexeus, Barcelona, Spain Institute of Medical Genetics and Applied Genomics, University of Tübingen, Tübingen, Germany

various genomic and evolutionary signatures. Clinical information, such as disease phenotype or mode of inheritance, is incorporated to improve the precision of the prioritization algorithm. Benchmarking against state-of-the-art competitors demonstrates that eDiVA consistently performed as a good or better than existing approach in terms of detection rate and precision. Moreover, we applied eDiVA to several familial disease cases to demonstrate its clinical applicability. only affects a small number of patients, the sum of people affected by one of the more than 7,000 rare diseases exceeds 30 million individuals in the US alone (Cutting, 2014). Whole-exome sequencing (WES), and more recently whole-genome sequencing (WGS), are routinely applied to identify variants causing rare Mendelian diseases in studies of families or parent-child trios (Choi et al., 2009;Ng et al., 2010;Louis-Dit-Picard et al., 2012;Rabbani, Mahdieh, Hosomichi, Nakaoka, & Inoue, 2012).
Usually, each exome sequencing experiment yields tens of thousands of genetic variants in coding and splicing regions that require thorough functional annotation and filtering to allow identification of the causal variant. Several tools have been published performing variant annotation, including Annovar, VEP, or SNPeff, which augment the sequencing information with a comprehensive set of current omics, population genomics, and clinical knowledge (Cingolani et al., 2012;McLaren et al., 2016;Wang, Li, & Hakonarson, 2010). These tools utilize a large selection of available databases containing gene annotations, various genomic features, variant allele frequencies in different populations, functional impact prediction, and evolutionary conservation (Bao et al., 2014). Other methods, such as eXtasy (Sifrim et al., 2013), PhenoDB (Sobreira, Schiettecatte, Boehm, Valle, & Hamosh, 2015), Phen-Gen (Javed, Agrawal, & Ng, 2014), VarSifter (Teer, Green, Mullikin, & Biesecker, 2012), KGGseq (M.-X. Li, Gui, Kwan, Bao, & Sham, 2012), and SPRING (Wu, Li, & Jiang, 2014), focus on prioritization of potentially causal variants using both functional annotation and clinical information. These tools systematically filter, evaluate, and prioritize thousands of variants, taking into account knowledge found in genome annotation databases (Rhead et al., 2010), disease gene repositories (OMIM, Online Mendelian Inheritance in Man; Landrum et al., 2014), and patient pedigree information, as well as phenotype descriptions and disease definitions provided for example, as Human Phenotype Ontology (HPO) terms . Finally, methods such as Endeavour (Tranchevent et al., 2008) and GeneDistiller (Seelow et al., 2008) prioritize disease genes, not individual variants, by integrating diverse genomic data sources.
Detection rates of causal variants using WES have been reported to be as low as 20-30% of cases (H. Lee et al., 2014;Yang et al., 2013), although higher success rates have been reported for specific disease or inheritance types (Sawyer et al., 2016) and for studies using parentchild trios (Yang et al., 2013). While some of the unsolved cases might be explained by intergenic or intronic regulatory variation or unidentified structural variants, the low detection rate also indicates the need for development of better prioritization strategies for coding variants and robust classifiers comprehensively integrating the available amount of prior omics and the knowledge of the disease.
Many computational algorithms have been developed to assess pathogenicity of genetic variants. Tools such as SIFT (Kumar, Henikoff, & Ng, 2009), CADD (Kircher et al., 2014), PolyPhen-2 (Adzhubei et al., 2010), or Eigen (Ionita-Laza, McCallum, Xu, & Buxbaum, 2016) are commonly used in clinical practice to help variant interpretation. They derive a functional impact score based on amino acid or nucleotide conservation, and biochemical properties of the amino acid changes as features. While some algorithms additionally categorize variants into various categories such as "neutral," "benign," "deleterious," "damaging," "probably-damaging," or "pathogenic" (e.g., SIFT, Condel, PolyPhen-2, and Mutation Assessor), scores of other methods need to be interpreted by using (often arbitrary) cutoffs for pathogenicity (e.g., CADD). These predicted pathogenicity labels are an integral part of the American College of Medical Genetics and Genomics standards and guidelines for the interpretation of sequence variants (Richards et al., 2015). Methods combining multiple classifiers, such as MetaLR, have been shown to produce better results than single classifiers (Dong et al., 2015).
Recently, specialized ensemble learning methods for estimating pathogenicity of rare variants have been published: Mendelian Clinically Applicable Pathogenicity (M-CAP; Jagadeesh et al., 2016), using gradient-boosting trees on pathogenicity scores and conservation features, and Revel (Ioannidis et al., 2016), using an RF to integrate several pathogenicity predictors.
To combine an intuitive user interface with comprehensive variant prediction, annotation, pathogenicity classification, and causal variant prioritization we developed eDiVA (exome Disease Variant Analysis), http://www.ediva.crg.eu. The eDiVA pipeline is composed of four main components: (a) eDiVA-Predict, where sequencing results are processed to predict the presence of genomic variants; (b) eDiVA-Annotate, that enriches variants via a domain-knowledge database; (c) eDiVA-Score, which estimates variant pathogenicity using a random forest model; and (d) eDiVA-Prioritize, in which variants from small groups of related samples (i.e., families or parent-child trios) are analyzed jointly. eDiVA returns a shortlist of candidate variants compatible with the selected disease inheritance model and the pedigree information. Using the pathogenicity probability computed by eDiVA-Score, variants are ranked such that better candidates appear on top of the result list. eDiVA has been developed with specific emphasis on usability, automation, and reproducibility of results and is available as a web service with a graphical user interface (see Supporting Information Material), or as an opensource repository with Docker containers. eDiVA can be run using the NextFlow (Di Tommaso et al., 2017) pipeline management system to ensure its compatibility with most standalone or cloud-computing platforms as well as to guarantee reproducibility on any system. eDiVA has been optimized for two common clinical diagnostics scenarios, parent-child trios comprised of healthy parents and one affected child (tested for recessive, compound heterozygous, and Xlinked inheritance or dominant de novo variants) and families with multiple affected relatives (additionally tested for dominant inheritance). We demonstrate that eDiVA outperforms competing approaches in a semisynthetic benchmark study introducing thousands of known disease variants from ClinVar (Landrum et al., 2014) or HGMD (Stenson et al., 2017) into real WES data from the 1000 Genomes Project CEPH parent-offspring trio of European ancestry (NA12878, NA12891, and NA12892). We, furthermore, report summary statistics on eDiVA and Phen-Gen results for 35 unreported disease cases, composed of 15 cases of spinocerebellar ataxia, 16 cases of primary immunodeficiency, and four cases of congenital myasthenia.  (Lek et al., 2016) databases. The latter three databases also provide information on specific populations (e.g., Caucasian, Asian, African American, etc.), which can be selected for improved causal variant prioritization. Information on evolutionary conservation is incorporated from PhyloP (Rhead et al., 2010), and PhastCons (Hubisz, Pollard, & Siepel, 2011), including conservation scores for primates, mammals, and vertebrates. Precalculated scores for functional impact of variants have been integrated based on the algorithms SIFT (Kumar, Henikoff, & Ng, 2009) | 867 by UCSC table browser (Rhead et al., 2010). Finally, eDiVA-DB provides clinical data from ClinVar (Landrum et al., 2014) and OMIM related to each variant and affected gene. eDiVA-Annotate uses multisample VCF files and returns a file with annotated variants in comma-separated value format. This step can be performed on the eDiVA web server.
2.4 | eDiVA-Score: Estimating variant pathogenicity eDiVA's prioritization algorithm relies on accurate estimation of pathogenicity for each variant. We therefore developed eDiVA-Score, a machine learning classifier, which assigns a pathogenicity probability to each variant based on its annotation characteristics obtained from eDiVA-Annotate. eDiVA-Score is built by training a random forest (RF) model using the R "randomForest" package with 1000 binary classification trees (Breiman, 2001;Hastie, Tibshirani, & Friedman, 2009) Figure S2a,b), thus circumventing that AF dominates the classification model. Following the neutral theory of molecular evolution (Kimura, 1983) missing data is generated using expected values for nonpathogenic (neutral) variants ( Figure 1). The only exception is AF, as missing data in the context of AF means that the SNV is novel, that is, has AF of zero. Variants used for training of the RF have been excluded in all benchmarking tests performed in this study.  Table S1. Finally, eDiVA allows the user to specify a list of HPO terms  relevant for the disease as an additional source of information to prioritize variants in genes. eDiVA highlights all variants in genes related to the submitted phenotypic traits using a custom algorithm to estimate the HPO-gene association (detailed in the Supporting Information material).

| Performance evaluation using semisynthetic cases
To assess the performance of eDiVA and several competing methods, we implemented a semisynthetic benchmark based on real WES data from a trio in which we spiked-in known pathologic variants from the ClinVar database (Landrum et al., 2014). We chose a publicly available CEPH trio sequenced within the framework of the 1000 Genomes Project composed of samples with European ancestry NA12878 (daughter), NA12891 and NA12892 (parents), downloadable from https://public_docs.crg.es/sossowski/MicrobeGenomes/human/eDiVA/ insilico_simulation_data/, and we called variants and generated a multisample VCF file using eDiVA-Predict. For the purpose of this benchmark study, all 138,705 variants found in the original trio are considered true negatives, that is, variants not associated with the disease.
Next, we embedded known disease variants in the trio following segregation patterns expected for Mendelian diseases. This positive set, containing variants associated with the diseases, consists of all variants from ClinVar (Landrum et al., 2014) database labeled as "pathogenic" or "likely pathogenic", having an OMIM reference in the database and that had not been used for training of eDiVA-Score. For each pathogenic variant, we extracted: chromosome, position, reference and alternative nucleotides, dbSNP identifier, gene name, inheritance mode of the associated disease (where available, randomly assigned otherwise), and HPO terms for the disease.
Variants without HPO annotation have been excluded from the benchmark set.
We have simulated three inheritance patterns: autosomal recessive homozygous, autosomal recessive compound heterozygous, and dominant de novo, as these are the most likely patterns found in parent-child trio based rare-disease diagnostics. To create realistic disease genotypes, each pathogenic variant was introduced into the exomes of the daughter and the parents, if applicable according to the inheritance mode. The read distribution of reference and alternative reads was simulated depending on the inheritance mode and the original coverage data. The variant allele frequency (VAF) of the alternative allele (i.e., the fraction of reads showing the alternative allele) introduced in the original VCF file has been obtained using a beta distribution and a binomial distribution for homozygous and heterozygous variants, respectively. A total of 6,811 disease-associated variants from ClinVar not previously used in the training of eDiVA-Score were used for benchmarking: 3,353 recessive homozygous, 2,592 dominant de novo, and 866 recessive compound heterozygous disease-causing variants (see Table S2 for additional information on simulated genotypes). We evaluated the ability of eDiVA-Score and six competing methods, namely CADD, Eigen, DANN, Revel, M-CAP, and MetaLR (Dong et al., 2015, Ioannidis et al., 2016, Ionita-Laza et al., 2016, Jagadeesh et al., 2016, Kircher et al., 2014, Quang, Chen, & Xie, 2015, to prioritize pathogenic over benign variants. We generated a receiver operating characteristic (ROC) curve for each tool and benchmark set and measured performance by area under the curve (AUC).
PhenoDB was executed from the https://phenodb.org/ website using standard parameters (a) AF, <0.01; (b) including variants which are present in dbSNP, and (c) analysis type chosen among "autosomal recessive compound heterozygous", "autosomal recessive homozygous", or "autosomal dominant new mutation". We locally installed Phen-Gen and launched it with the corresponding setups: (a) "Recessive", "allow_de_novo = 0" for recessive and compound inheritance, and (b) "Dominant", "allow_de_novo = 1" for the dominant de novo inheritance model. We locally installed Exomiser and analyzed all trio cases using PhenIX prioritization mode (details in Supporting Information Material). We tested eDiVA in two configurations, (a) without phenotype description, and (b) using HPO IDs describing the disease phenotype for disease-specific prioritization of candidate variants.
To benchmark the ability of eDiVA, Exomiser, PhenoDB, and Phen-Gen to distinguish disease-causing from benign variants we

| Benchmarking eDiVA and competing methods
To comprehensively evaluate eDiVA's performance in finding disease-causing variants, and to compare it to previously published tools, we performed a benchmark in two categories. First, we

| Benchmarking of eDiVA-Score and other variant pathogenicity classifiers
We developed eDiVA-Score, a machine learning-based method for estimating variant pathogenicity (deleteriousness) independent of any prior clinical information (see Section 2). Feature-selection identified population allele frequency, functional impact, and conservation in placental mammals as the most important features ( Figure 1a). The correlation matrix for all features is shown in Figure S3. Features selected for inclusion in the RF show distinct distributions for pathogenic variants compared to benign variants in ClinVar (Figure 1b), random coding variants reported in GnomAD ( Figure S4b). All integrated conservation scores (PhyloP and PhastCons scores for vertebrates, mammals and primates) classify pathogenic variants better than random, but perform worse than any specialized method for estimating functional impact or pathogenicity ( Figure S5).
We benchmarked the ability of eDiVA-Score, CADD, DANN, Eigen, Revel, M-CAP, and MetaLR to predict the deleteriousness of variants and to distinguish pathogenic from benign variants in nine setups (Section 2). We first compared the performance on classifying pathogenic and benign variants from ClinVar (Figure 2a), on distinguishing disease variants from HGMD (Stenson et al., 2017) from 100,000 random variants from GnomAD (Figure 2d), for which scores are available for all methods. Note that Revel and M-CAP have been trained on a subset of the HGMD variants (e.g., using class "DM" as positive training set), giving them an advantage due to potential overfitting in any of the following benchmark tests using HGMD variants (for an in-depth discussion of the interplay between overfitting and circularity in training and benchmarking data (Grimm et al., 2015). Using ROC analysis, we found that eDiVA-Score values for any position of the exome, since de novo mutations can occur randomly and novel ultra-rare variants are still being discovered. Therefore, we next benchmarked the methods on random variants chosen from the complete ClinVar and HGMD/ GnomAD benchmark sets, whereas setting missing data to benign (Section 2). As expected, the recall of Revel, M-CAP, and MetaLR decreased substantially due to missing information, whereas the other methods performed slightly better than in the previous tests (Figure 2b,e). Finally, we tested how the methods perform on classification of rare variants (AF, <0.01), otherwise following the same criteria for selection of the test sets as in the previous benchmark (Figure 2c,f). Again, eDiVA-Score shows the best performance of all methods.
We wondered if the use of random GnomAD variants as TN (nonpathogenic) set might bias the results of the HGMD/GnomAD benchmark due to for example, overfitting onto the allele frequency feature. Therefore, we next measured the performance of all methods on HGMD data only, using the categories for highly likely pathogenic ("DM" and "DM?") as TP set and less likely pathogenic (any other HGMD category) as TN set (Section 2). We performed the same three tests as discussed above for the ClinVar and HGMD/ GnomAD benchmark sets. On the subset of variants for which scores are available for all methods (Figure 2g Finally, we compared the performance of all methods on a benchmark set compiled by Grimm et al. (2015), consisting of mutually exclusive subsets of the previously published benchmark sets Varibench, HumVar, ExoVar, predictSNP, and SwissVar (see Supporting Information Material for details). These popular benchmark data sets differ in the way they define pathogenic and neutral variants, for example, the maximum AF for pathogenic variants can differ dramatically, allowing us to benchmark diverse challenges.
Furthermore, Grimm et al. filtered these benchmark sets to minimize overlap between them, reducing the likelihood that tools are benchmarked on variants they have been trained on and hence reducing the impact of overfitting on the benchmark results (Grimm et al., 2015). We found that none of the methods consistently performs better than other methods, but that eDiVA-Score, M-CAP, Revel, and MetaLR show comparably high performance, except on PredictSNP and Varibench, for which MetaLR, Revel, and M-CAP show a better performance than eDiVA-Score. PredictSNP incorporates HGMD variants in the positive and negative control sets; see Tables 2 and 3 of Grimm et al. (2015), likely to be giving a strong advantage to Revel and M-CAP, which have been trained on HGMD.
CADD, DANN, and Eigen performed significantly worse than the other three methods on all benchmark sets. Note that CADD, DANN, Eigen, and MetaLR have been trained to predict deleteriousness (or more general the functional impact) of variants, whereas eDiVA-Score, Revel, and M-CAP have been trained to identify pathogenic variants, partly explaining the divergent performance levels across the different benchmark sets. Moreover, eDiVA-score, MetaLR, and M-CAP use CADD as one of many features, explaining the better performance of the derived scores.
In summary, our benchmark results demonstrate the good performance of eDiVA-Score as pathogenicity classifier, comparable to and often better than state-of-the-art methods available to date.

| Causal variant prioritization in parent-child trios
We benchmarked the performance of eDiVA and three widely used tools, PhenoDB, Phen-Gen, and Exomiser, on identification of causal variants for rare Mendelian diseases in parent-child trios. To this end, we simulated three scenarios typically encountered in parentchild trio diagnostics, (a) autosomal dominant de novo, (b) autosomal recessive homozygous, and (c) autosomal recessive compound heterozygous Mendelian inheritance modes. In total, we simulated 6,811 semisynthetic parent-child trios by integrating reported pathogenic variants from ClinVar into real WES data of a trio obtained from 1000GP (see Section 2 and Table S2).

| eDiVA results on clinical cases
eDiVA has successfully been used in published case studies on mitral valve prolapse (Durst et al., 2015), cystic fibrosis (Ramos et al., 2014), phenylketonuria , arthrogryposis (Wambach et al., 2017), and Opitz-C (Urreizti et al., 2017), among others, identifying both known, as well as novel rare-disease genes. We recently assessed the performance of eDiVA for the diagnosis of rare congenital genetic diseases using WES of 35 parent-child trios, including 15 cases of congenital ataxia, four cases of congenital myasthenia, and 16 cases of primary immunodeficiency. Here we report general statistics on the number of candidate genes per case, while case reports and novel candidate genes will be published in separate papers. Across the 35 studied trios, eDiVA on average reported a median of 11, 3, and 10 candidate genes per trio for recessive homozygous, recessive compound heterozygous, and dominant de novo inheritance mode, respectively, using default parameters. In comparison, Phen-Gen reported a median of 36 and 52 candidate genes for recessive (including compound heterozygous) and dominant (including de novo) inheritance mode, respectively.

Histograms of reported candidate gene numbers for eDiVA and
Phen-Gen are shown in Figures S9 and S10. eDiVA found causal variants in known genes for the respective disease in none cases, and variants in genes associated with closely related disease phenotypes in seven cases. Screening of Phen-Gen results did not reveal additional candidates missed by eDiVA. The function of a novel disease variant for congenital ataxia has been described in Bahamonde et al. (2015), and reports for other candidate genes are in preparation.

| DISCUSSION
Despite the massive increase in sequencing capacity and the availability of highly optimized analysis tools, multiple large-scale rare-disease studies reported that in only 20-30% of cases a causal variant can be identified using WES. Several reasons might explain the inability of WES analysis to identify causal variants in a majority of cases, including for example, the inability to identify regulatory variants (Claussnitzer et al., 2015), our limited knowledge of the function of noncoding RNAs, generation of new exon donor or acceptor sites by intronic variants (Y. Lee et al., 2012), small copy number variations (Krumm et al., 2012), incomplete penetrance, and unknown function of coding genes, among others. However, we argue that the potential of WES has not been exhausted and that causal coding variants are often missed due to inappropriate correction of noise in the data, insufficient use of clinical (phenotypic) data, or reporting of long unranked candidate lists, requiring tedious screening by clinicians. We further claim that these shortcomings are often overlooked due to unrealistic simulated benchmark tests not reflecting the level of noise found in real family or trio NGS data.
We have addressed these problems by developing eDiVA, a pipeline that combines multisample variant calling of family data, QC and filtering, extensive functional annotation, machine learningbased classification of deleterious variants, and prioritization of causal variants optimized for various clinical scenarios. Furthermore, we developed a highly realistic benchmark test combining real WES data of a parent-child trio with thousands of pathogenic ClinVar variants to generate 6,811 semisynthetic disease trios. Using these cases, we have demonstrated that eDiVA's pathogenicity estimator (eDiVA-Score) as well as eDiVA's prioritization algorithm perform favorably compared to existing state-of-the-art methods. eDiVA has been able to find disease-causing variants with higher recall, fewer false positives and better ranking than competing tools in three benchmarked modes of inheritance. Finally, we evaluated the use of phenotypic descriptors for optimizing the prioritization process.
We found that adding HPO ID-based prioritization introduces a trade-off between recall and the number of false positives in the output list. Despite the marginal reduction in recall, focusing on known disease genes is often the preferred choice for diagnostic purposes. Our knowledge of genetic factors playing a role in disease is constantly growing, reflected in a rapid increase of genotypephenotype relations stored in various databases. Hence, it would be beneficial to reanalyze WES data sets once in a while (e.g., every 6-12 months) to benefit from new knowledge and to facilitate identification of previously unknown/unreported causal variants.
Moreover, combined reanalysis of the growing cohorts of WES data stored in many institutes would allow to identify matching causal genes across multiple families or cases. However, most analysis pipelines require substantial hands-on time and long candidatevariant lists have to be screened by experts, making regular reanalysis of data sets impractical. eDiVA has been developed with a specific emphasis on high reproducibility of results and complete automation of the analysis using artificial intelligence-based methods.
Machine learning classifiers are used to perform candidate ranking and prioritization, reducing hands-on time of clinical experts to a minimum. Integration with NextFlow, moreover, guarantees reproducibility of results at later time points and on most computing platforms. Therefore, eDiVA is a dedicated solution for regular reanalysis of large disease cohorts or collections of diagnostic cases.
Additional steps can be taken to improve the identification of disease-related variants from WES data. The availability of custom allele frequency databases with geographical specificity would help to reduce the number of false-positive genotype-phenotype associations due to population specific variants. To this end, institutes and hospitals with access to large cohorts of sequenced exomes may use in-house data to filter population specific variants, an approach we have pursued our self by collecting thousands of Iberian cases in an aggregated allele frequency database (http://geevs.crg.eu/, unpublished). Identification of extended homozygosity regions could in addition help to diagnose causal homozygous variants in consanguineous cases. Moreover, the integration of structural and copy number variants (SVs and CNVs) has BOSIO ET AL. | 875 been shown to increase recall rates substantially (Gambin et al., 2017).
Despite their frequent involvement in rare diseases (McCarroll & Altshuler, 2007), CNVs are often disregarded in WES analyses, and are rarely processed in combination with point mutations. Prioritization algorithms will have to be adapted to consider compound heterozygotes composed of a point mutation in one and a CNV in the other allele. CNV analysis is currently being integrated in eDiVA and will be available in the near future.
Better use of phenotypic descriptors has the potential to improve both precision and recall of causal variant prioritization methods. We observed that HPO ID-based prioritization dramatically improved the precision of eDiVA. However, incomplete maps of known genotype-phenotype (or gene-phenotype) relations in public databases led to a mild reduction in recall. Robinson et al. (2014) proposed a method to overcome this limitation, tapping into the genotype-phenotype associations from mouse data to solve causal variant identification for corresponding human phenotypes. Other methods based on image analysis, for example, Hadj-Rabia et al.
(2017) or face2gene (http://suite.face2gene.com/), have also shown promising results for diagnosis of patients with visible phenotypic features. Finally, an important step in the evaluation of newly discovered genotype-phenotype associations is the identification of additional cases with a similar phenotype and mutations in the same gene. Several approaches for gene matching have been published, for example, GeneMatcher (Sobreira et al., 2015), which have been connected via the Matchmaker Exchange platform. Integration of approaches using image analysis, cross-species phenotype-genotype correlation, and gene matching has the potential to further improve AI-based variant prioritization methods such that they can rival the diagnostic precision of clinical experts in the future.
In summary, we have shown that eDiVA is a step towards full automation of causal variant identification in family and parent-child trio data using machine learning-based approaches. eDiVA can be used as a support tool for clinicians to find disease-causing variants, or as a fully automated solution for periodic reanalysis of large WES (or WGS) cohorts. eDiVA is able to identify known causal disease variants with high precision and recall, and facilitates identification of novel disease variants with minimal hands-on time.