GENIE: Genetic Evaluation and Naive Inference for Early Diagnosis of Fabry Disease

Abstract

Fabry Disease is a rare lysosomal disorder that reduces the body's ability to decompose glycosphingolipids that naturally accumulate in lysosomes. Specifically, the disease involves mutations in the galactosidase alpha (GLA) gene, preventing adequate production of the enzyme a-galactosidase (a-Gal). This enzyme is responsible for the breakdown of glycosphingolipids, which if not metabolized, can harm the involuntary functions of the nervous and cardiovascular systems, eyes, and kidneys. To date, the only known Fabry Disease treatment is 1-deoxygalactonojirimycin, otherwise known as oral migalastat. Fabry Disease typically falls under two categories: Classic and Late-Onset. The former develops during childhood or adolescence, while the latter is not evident until early adulthood. Current diagnosis methods are tedious and time-consuming, and the disease may spread tremendously before being identified and treated. To address this inefficiency, two machine-learning classifiers were developed: one Linear model and one Naive Bayes model. Given a mutated a-GLA nucleotide sequence, each model was to determine the mutation's Fabry variant and its amenability to oral migalastat. The Linear classifier achieved an accuracy of 59%, while the Naive Bayes classifier reached an accuracy of 96%. Thus, a correlation was established between the independent features of a Fabry-affected genotype and the patient's phenotype—an observation that will tremendously improve Fabry Disease diagnosis and treatment.