PON-P3 is a computational method for interpreting the potential pathogenicity of amino acid substitutions. Its development combines carefully selected variant datasets, feature selection and an ensemble prediction strategy. Starting from 1,526 candidate features describing variants, sequence and structural context, genes and proteins, the study evaluates the resulting method against 23 other predictors. Unlike binary classifiers, PON-P3 can assign variants to benign, pathogenic or uncertain-significance categories. The method was applied to substitutions in human proteins defined by MANE transcripts and to variants of uncertain significance in ClinVar. This work connects statistical methodology with molecular and genetic evidence while explicitly retaining uncertainty in variant interpretation.