Structure-informed classification of RyR1 variants highlights limitations of current predictors and enables clinical interpretation

RyR1-related disorders, arising from variants in the RYR1 gene encoding the skeletal muscle ryanodine receptor, encompass a wide range of dominant and recessive phenotypes. The extensive length of RyR1 and diverse mechanisms underlying disease variants pose significant challenges for clinical interpretation, exacerbated by the limited performance and biases of current variant effect predictors (VEPs). This study evaluates the efficacy of 70 VEPs for distinguishing pathogenic RyR1 missense variants from putatively benign variants derived from population databases. Existing VEPs show variable performance. Those trained on known clinical labels show greater classification performance, but this is likely heavily inflated by data circularity. In contrast, VEPs using methodologies that avoid or minimise training bias show limited performance, likely reflecting difficulty in identifying gain-of-function variants. Leveraging protein structural information, we introduce Spatial Proximity to Disease Variants (SPDV), a novel metric based solely on three-dimensional clustering of pathogenic mutations. We determine ACMG/AMP PP3/BP4 classification thresholds for our method and top-performing VEPs, allowing us to assign PP3/BP4 evidence levels to all available RyR1 missense VUSs in ClinVar. Thus, we suggest that our protein-structure based approach represents an orthogonal strategy over existing computational tools for aiding in the diagnosis of RyR1-related diseases.