Large, poorly resolved human proteins remain a major barrier to precision medicine because their size and structural complexity hinder accurate interpretation of genetic variants. The ryanodine receptor 1 (RyR1) is a 2.3 MDa homotetrameric calcium-release channel that drives skeletal muscle contraction and is a frequent site of disease-causing mutations. Variants in RyR1 underlie rare congenital myopathies, yet many remain classified as variants of uncertain significance, limiting diagnosis and patient care.
Here we present the first full-length human RyR1 structural model and, leveraging it, the first dedicated structure-aware machine-learning classifier for RyR1 variant pathogenicity. The structure was built by multi-template homology modelling guided by high-resolution rabbit cryo-EM data and assembled into a tetrameric channel. From this model we extracted a rich dataset of sequence- and structure-informed descriptors including evolutionary conservation, physicochemical change, pocket geometry, interaction networks, energetic (de)stabilisation and disorder propensity to train the predictor.
Our model markedly outperforms widely used tools such as SIFT, PolyPhen-2, ESM1b and AlphaMissense (with an MCC = 0.84) and provides residue-level mechanistic insight into how mutations perturb channel stability and function. By bridging structural biology and machine learning, this platform enables classification of both novel and previously reported uncertain variants, supporting precision genetic diagnostics and laying the groundwork for future targeted therapies in RyR1-related disease.