Integrating Functional Consequence Annotation With PAH Allelic Phenotype Values Refines Prediction of Tetrahydrobiopterin Responsiveness

Tetrahydrobiopterin (BH4; sapropterin) responsiveness in phenylalanine hydroxylase (PAH) deficiency is genotype dependent, yet many patients remain untested. Allelic phenotype values (APV) summarize allele severity, but responsiveness can be heterogeneous within APV strata. We assessed whether integrating functional consequence annotation from Ensembl variant effect predictor (VEP) improves genotype-based prediction of BH4 response. We analyzed 23 640 individuals with biallelic PAH genotypes and BH4 status (RESP, S-RESP, N-RESP, or not tested). Tested individuals were used for model development (responders defined as RESP+S-RESP; nonresponders as N-RESP). APV values were assigned from published APV resources and merged at the variant level. Functional consequence predictors were derived from VEP output and included strict predicted loss-of-function (pLoF) flags, splice-impact scores (SpliceAI maximum delta score), and missense pathogenicity predictions (SIFT and PolyPhen). Genotype predictors were constructed using a milder-versus-severer allele framework, with the milder allele defined as the allele with the higher APV. Models were evaluated using genotype-held-out cross-validation (GroupKFold by genotype). Among 4640 tested individuals, 2044 (44.1%) were BH4 responders. Responder rates were enriched in milder phenotypes and increased monotonically across milder-allele APV bins. In genotype-held-out evaluation, integrating functional consequence predictors with APV improved discrimination modestly overall and more clearly in intermediate APV genotypes. Genotype predicts BH4 responsiveness with high performance under stringent genotype-held-out validation, and VEP-derived functional consequence annotation provides modest complementary value beyond APV, particularly for intermediate-severity genotypes.