Autism spectrum disorder (ASD) is a neurodevelopmental disorder that lacks clear biological biomarkers. Existing diagnostic methods focus on behavioral and performance characteristics, which complicates the diagnosis of patients younger than 3 years-old. The purpose of this study is to characterize metabolic features of ASD that could be used to identify potential biomarkers for diagnosis and exploration of ASD etiology. We used gas chromatography-mass spectrometry (GC/MS) to evaluate major metabolic fluctuations in 76 organic acids present in urine from 156 children with ASD and from 64 non-autistic children. Three algorithms, Partial Least Squares-Discriminant Analysis (PLS-DA), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost), were used to develop models to distinguish ASD from typically developing (TD) children and to detect potential biomarkers. In an independent testing set, full model of XGBoost with all 76 acids achieved an AUR of 0.94, while reduced model with top 20 acids discovered by voting from these three algorithms achieved 0.93 and represent a good collection of potential ASD biomarkers. In summary, urine organic acids detection with GC/MS combined with XGBoost algorithm could represent a novel and accurate strategy for diagnosis of autism and the discovered potential biomarkers could be valuable for future research on the pathogenesis of autism and possible interventions.
Chen, Q., Qiao, Y., Xu, X. J., Tao, Y., & You, X. (2019). Urine organic acids as potential biomarkers for autism-spectrum disorder in chinese children. Frontiers in Cellular Neuroscience, 13. https://doi.org/10.3389/fncel.2019.00150