Diagnostic yield and clinical utility of whole exome sequencing using an automated variant prioritization system, EVIDENCE

Abstract EVIDENCE, an automated variant prioritization system, has been developed to facilitate whole exome sequencing analyses. This study investigated the diagnostic yield of EVIDENCE in patients with suspected genetic disorders. DNA from 330 probands (age range, 0‐68 years) with suspected genetic disorders were subjected to whole exome sequencing. Candidate variants were identified by EVIDENCE and confirmed by testing family members and/or clinical reassessments. EVIDENCE reported a total 228 variants in 200 (60.6%) of the 330 probands. The average number of organs involved per patient was 4.5 ± 5.0. After clinical reassessment and/or family member testing, 167 variants were identified in 141 probands (42.7%), including 105 novel variants. These variants were confirmed as being responsible for 121 genetic disorders. A total of 103 (61.7%) of the 167 variants in 95 patients were classified as pathogenic or probably to be pathogenic before, and 161 (96.4%) variants in 137 patients (41.5%) after, clinical assessment and/or family member testing. Factor associated with a variant being regarded as causative includes similar symptom scores of a gene variant to the phenotype of the patient. This new, automated variant interpretation system facilitated the diagnosis of various genetic diseases with a 42.7% diagnostic yield.

Patients aged ≥5 months were included if they were strongly suspected by clinicians of having a genetic disease and were undiagnosed, despite application of conventional genetic tests, such as, chromosome analyses, chromosome microarray, or single or targeted gene panel testing. Patients aged <5 months were included if they had a congenital anomaly in one or more major organs, including the brain or the heart; or the gastrointestinal, urological, or musculoskeletal systems; or if they were strongly suspected by clinicians or radiologists of having a genetic disease.

| Whole exome sequencing, variant calling, and variant annotation
Blood, saliva, or buccal swab samples were collected from each patient, and genomic DNA was extracted from each sample. All exon regions of all human genes (~22 000) were captured using the Agilent SureSelect kits (version C2, December 2018) and sequenced using the NovaSeq platform (Illumina, San Diego, CA). The quality of FASTQ files obtained by sequencing with the Illumina Novaseq 6000 was assessed using FASTQC (http://www.bioinformatics.babraham.ac.uk/ projects/fastqc/). Subsequently, the base and sequence adapters with low base quality were removed using Trimmomatic. 18

| EVIDENCE: Prioritization of variants and symptom suggestion system
The streamlined variant prioritization software program, EVIDENCE, was developed in-house to prioritize variants based on ACMG guideline and the phenotype of each patient and to interpret these variants accurately and consistently. This system has three major steps: variant filtration, classification, and similarity scoring for patient phenotype. In the first step, allele frequency was estimated in population genome databases, including gnomAD (http://gnomad.broadinstitute.org/) and 3billion Inc. (https://3billion.io/). 22 Common variants with a minor allele frequency of >5% were filtered out in accordance with rule BA1 of the ACMG guidelines. 17 In the second step, Evidence of the pathogenicity of the variants was obtained from disease databases, including OMIM (www.omim. org), ClinVar, and UniProt; the factors included gene function, domain of interest, mechanism of development, inheritance pattern, and clinical relevance of the disease. 1,23,24 The predicted functional or splicing effect of each variant and its degree of evolutionary conservation was evaluated using several in silico tools, including REVEL, adaptive boosting, and random forest score. 25,26 Scores above 0.5 in each tool predicted a detrimental effect on the variant. The pathogenicity of each variant was evaluated according to the recommendations of the ACMG guidelines. 17 In the third step, the clinical phenotype of each proband was transformed to its corresponding standardized HPO term and was assessed to measure the similarity with each of~7000 rare genetic diseases. 27,28 The similarity score between the phenotype of each patient and symptoms associated with that disease, caused by prioritized variants, according to ACMG guidelines, ranged from 0 to 10. For any given symptom in a patient, the symptom was compared with each of the known symptoms of a target disease. At each comparison of two symptoms, the maximal depth of a common ancestor node of two symptoms was registered as a weight candidate. As the given symptom of the patient was compared with each of the disease symptoms, the number of weight candidates and the number of disease symptoms were the same. The weight of a symptom was set as the maximum value of the registered candidate weights. Weights were calculated for all patient symptoms and averaged to S1. Using the same procedure, all known symptoms of the target disease were weighted and averaged to S2. The value, (S1 + S2)/2, was used as the score for symptom similarity between the patient and the target disease. The formula used to calculate the similarity score is (supplementary file 1): Incidental findings were not included in this study. Finally, EVI-DENCE prioritized variants that were classified as pathogenic, probably pathogenic, or VUS according to ACMG guidelines, were categorized into a three tier system based on the Bayesian score. 29 The first tier was scored above 0.9, the second tier above 0.499, and the third tier above 0.1. These variants were ranked higher as the similarity score within each tier was high. A diagram highlighting each step of the filtering process used for variants and databases related to this process are presented in Supplementary Figure 1 and Table S1, respectively.
The entire process of genetic diagnosis, including processing of raw genome data, determining variant prioritization, and measuring the similarity between each phenotype and disease, was integrated and automated into a computational framework.

| Variant interpretation and confirmation
Relevant candidate variants, including VUS, based on EVIDENCE, were manually reviewed, related to applied ACMG rules and disease characteristics, and then selected by medical geneticists. After another examination in the outpatient clinic, the DNA of each patient and/or their parents was subjected to Sanger sequencing to confirm the candidate variant(s). If necessary, a chromosomal microarray was performed to assess uniparental disomy (UPD) or structural variants after Sanger sequencing of the family members.

| Statistical analysis
All statistical analyses were performed using the R studio software Eigen C++ linear algebra library, with P < .05 considered statistically significant.

| Patient demographics
The demographic characteristics of the 330 patients have been shown in Table 1. Mean ages at clinical presentation and when WES was performed were 5.9 ± 12.9 years (range, 0-68 years) and 11.9

| Diagnostic yield and classification of identified variants
The number of patients with variants and the identity of these variants have been summarized in Figure 1.

| Identification of ultra-rare genetic disorders and its impact on clinical management
An ultra-rare genetic disease is defined as affecting 1 < 50 000 individuals. 31 In this study, 91 (75.2%) of 121 genetic disorders identified were ultra-rare genetic disease (Table S1).

| DISCUSSION
EVIDENCE, an automated variant prioritization system, was found to be useful in the entire WES process, including raw data processing, variant prioritization, and measurement of phenotypic similarity between patients and suggested candidate diseases.
The diagnostic yield of EVIDENCE in the present study (42.7%) was comparable with that previously reported for automated systems (30-35%). 9,15,16 This finding was important, as the phenotypes of the enrolled patients were quite heterogeneous, broadly dispersed, and not limited to certain organ categories. In addition, most of the diseases identified (75.2%) were ultra-rare genetic diseases. The distribution of inheritance patterns in the identified genetic disorders was generally similar to those in these reported studies, except that the proportion of autosomal dominant disorders was higher in our study. 6,8,10 Diagnosis rates over 40% have been reported in the absence of an automated system in patients with select disease phenotypes, including hearing loss, visual impairment, or abnormalities of the musculoskeletal system, as well as in patients with critical conditions and in new-borns presenting with symptoms. 4,8,32 Moreover, in the absence of an automated system, a large amount of time is required to interpret a significant number of variants in each patient. 6,11 The results presented here indicate that our automated variant prioritization system can contribute to diagnosing various types of genetic diseases with comparable accuracy, but with much greater speed, than non-automated analyses.
The comparable rate of diagnosis achieved by the automated system may be due to its high-performance efficiency. Based on the sys- Another factor responsible for the high diagnostic rate of this automated system was that a substantial proportion of the variants suggested by the system were VUSs. These VUSs were subsequently tested in family member segregation analysis and phenotype reassessment, as it is unclear whether VUSs are causative variants in the absence of segregation analysis and clinical reassessment.

Updated information on variants in genome databases can result in
VUSs being classified as pathogenic or benign. 12 In conclusion, the rate of detection of variants by the automated system did not differ significantly in patients who did and did not undergo genetic testing before WES. This automated system achieved a comparable diagnostic yield in patients with a broad range of genetic diseases, suggesting that WES may be one of the first diagnostic methods used in patients suspected of having a genetic disease, and that the automated system can facilitate the diagnostic process. This new method is available to others (https://portal.3billion.io/) allowing the efficiency of this system to be evaluated by other groups for larger patient cohorts. Phenotype-centric tools, such as, Phenovar or Exomiser have been recently developed. 9,15,16 EVIDENCE also uses a phenotype-centric approach, but prioritized variants are ranked in order in EVIDENCE (but not in the other systems). Furthermore, we are currently developing an ungraded system that prioritizes variants by merging the variant classification and similarity score into a single system, which will improve the analytical methods used to evaluate variants by EVIDENCE.