Evaluating the utility of REVEL and CADD for interpreting variants in amyotrophic lateral sclerosis genes

Abstract

Amyotrophic lateral sclerosis (ALS) is a debilitating neurodegenerative disease affecting approximately two per 100,000 individuals globally. While there are many benefits to offering early genetic testing to people with ALS, this has also led to an increase in the yield of novel variants of uncertain significance in ALS-associated genes. Computational (in silico) predictors, including REVEL and CADD, are widely employed to provide supporting evidence of pathogenicity for variants in conjunction with clinical, molecular, and other genetic evidence. However, in silico predictors are developed to be broadly applied across the human genome; thus, their ability to evaluate the consequences of variation in ALS-associated genes remains unclear. To resolve this ambiguity, we surveyed 20 definitive and moderate ClinGen defined ALS-associated genes from two large, open access ALS sequencing datasets (total people with ALS = 8,230; controls = 9,671) to investigate REVEL and CADDs ability to predict which variants are most likely to be disease-causing in ALS. While our results indicate a predetermined pathogenicity threshold for REVEL that could be of clinical value for classifying variants in ALS-associated genes, an accurate threshold was not evident for CADD, and both in silico predictors were of limited value for resolving which variants of uncertain significance (VUS) may be likely pathogenic in ALS. Our findings allow us to provide important recommendations of the use of REVEL and CADD scores for variants, and indicate that both tools should be used with caution when attempting to evaluate the pathogenicity of VUSs in ALS genetic testing.