Phenotype Correlations of Neurological Manifestations in Wolfram Syndrome: Predictive Modeling in a Spanish Cohort

Background: Wolfram syndrome (WS) is an ultrarrare genetic disorder caused by pathogenic variants in the WFS1 gene, combining endocrine and neurological involvement, leading to progressive neurological, autonomic, and cognitive impairment. Predicting neurological progression remains a clinical challenge, particularly in relation to genotype. Methods: Forty-five genetically confirmed patients with WS were followed in Spain between 1998 and 2024. Genetic variants were classified by exon, zygosity, and predicted wolframin production (Classes 0–3). Machine learning algorithms, including Random Forest models with gene–gene interaction terms, were applied to identify the strongest predictors of neurological involvement and to stratify phenotypic severity. Results: The most prevalent neurological signs were absence of gag reflex (67%), gait instability (64%), and dysphagia (60%), typically emerging in the third decade of life Homozygosity for truncating variants—especially c. 409₄24dup16 (Val142fsX110) —with no wolframin protein (Class 0) were the main predictors of early and severe neurological impairment. Machine learning models achieved accuracies between 88. 9% and 93. 3%, with wolframin class and allele 2 mutation ranking as top features. Conclusions: Integrating genetic and clinical data through machine learning enables robust prediction of neurological outcomes in WS. This approach enhances precision diagnosis and provides a framework for individualized monitoring of rare neuroendocrine disorders.