Urinary Metabolomics and One-Class Classification To Discover Children Affected by Bile Acid Synthetic Disorders: A Case-Study

Bile acid synthetic disorders (BASDs) are a group of rare metabolic disorders that involve the synthesis of bile acids. Symptoms often overlap with numerous other liver disorders, and an early diagnosis in childhood is extremely important. 3β-Dehydrogenase deficiency (BASD1), 5β-reductase deficiency (BASD2), and 27-hydroxylase deficiency, known as cerebrotendinous xanthomatosis (CTX), are probably the most common forms of BASD. The aim of the present study was to develop a machine learning based on one-class classification (OCC) capable of distinguishing subjects affected by BASD1 or BASD2 or CTX from patients presenting with altered liver function tests due to other causes, considering the urinary metabolome quantified by mass spectrometry coupled to liquid chromatography (LC-MS). Specifically, targeted metabolomics data were modeled by a multivariate one-class classifier that combines principal component analysis (PCA) and K-nearest neighbors (KNN) within the framework of model population analysis (MPA). To improve the performance of the classifier, the original set of quantified metabolites used in our research group was expanded, including the compounds discovered by an untargeted metabolomics investigation. Moreover, the strategy implemented for model interpretation allowed the identification of sets of metabolites closely related to the type of BASD. The proposed approach may be applied to other rare diseases.