Evaluating ACMG/ClinGen PP3/BP4 recommendations for missense variants in CSNK2A1

Purpose: Assessing variant pathogenicity in genes related to rare genetic disorders is a challenging task. While populational databases aid, additional methods are imperative when those genes are also constrained against variation, i.e. many potential variants are also absent from population databases. Many computational prediction algorithms (in silico tools) have been developed considering the protein and amino acid characteristics, and cross species conservation for assessing a variant pathogenicity. Some of those in silico tools are widely utilized by clinical and molecular geneticists and endorsed by professional organizations such as ACMG and ClinGen. However, their performance may not be the same on every gene and their variants. Materials and methods: In this study, the performance characteristics of ACMG/ClinGen endorsed in silico tools for pathogenic/likely pathogenic (reported in affected individuals) and benign/likely benign (high population allele frequency) missense variants in CSNK2A1 are evaluated to identify the most reliable prediction tool(s) in aiding variant pathogenicity assessment. Results: Among the endorsed in silico tools AlphaMissense is the best predictor for variant pathogenicity followed by MutPred2, VARITY_R, and ESM1b; while REVEL, VEST4, and BayesDel do not seem to be good predictors for PP3. Conversely, REVEL and BayesDel are the most reliable predictors for variant benignity compared to the rest of the predictors. Conclusion: Although the diagnostic laboratories are recommended to select one in silico predictor to utilize genome-wide variant predictions, every gene might benefit their own in silico predictor evaluations, even different predictors for pathogenic vs benign predictions might be better utilized.