Fig. 3

Variant analysis of patients with rare diseases. Panel A Overview of the filtering steps and the number of variants in rare disease patients referred for WGS analysis (means of 6 patients). The total number of variants in each patient is just above 5 MIO. The analysis begins by elimination of ~ 200.000 low quality variants. Subsequently, common variants with an allele frequency above 2% are excluded, since these are considered unlikely to explain the occurrence of a rare disease. Known pathogenic variants are retained. Since gnomAD may not represent all common variants, variants are moreover filtered against a local (Danish) reference genome and this further reduces the number of variants to about 200.000. Thereafter, the analysis is focused on coding and splice site variants and on average this reduces the number of variants to ~ 2400. Application of additional filters e.g., omitting ACMG/AMP benign variants or those with low REVEL scores further brings the number of variants down to ~ 1500. Panel B On average the patients exhibit 83 loss of function (LOF) variants and 748 missense variants. The remaining variants belonged to other categories such as variants in the UTRs and deep into the intron. Finally, on average 67 variants were previously registered in ClinVar or HGMD and information on these can be readily retrieved and used in the interpretation. The pie chart below shows the ACMG/AMP classification of the variants showing that only a minority are classified as pathogenic and likely pathogenic (< 2.5%). On average only a single pathogenic variant is identified. In many cases the variant represents a recessive heterozygote variant with no obvious relevance for the patient’s disease. Almost one third of the variants represents variants of unknown significance (VUS). Panels C and D shows the total cumulative distribution of gnomad allele frequencies and REVEL scores of ACMG/AMP scored variants (from Varseq) among 63 unrelated patients, respectively. Intergenic variants were filtered away and any variant which had conflicting classifications was removed. Moreover, variants with an allele frequency of more than 0.5 or for which an allele frequency could not be found was removed. The results illustrate that allele frequency is relatively effective in excluding benign variants, whereas likely benign and VUS are not effectively separated from the likely pathogenic and pathogenic variants by frequency filtering. The REVEL score combining pathogenicity predictions from 18 individual scores, in contrast, is clearly discriminative and high scores are enriched among pathogenic variants. About 25% of the VUS exhibit REVEL score above 0.5 that may warrant further analysis of these variants. The number and details of variants in the plots is summarized the attached Supplemental data