Machine Learning of Urinary Essential Element Concentrations from Children with Autism Spectrum Disorder and their Mothers

Introduction: Autism spectrum disorder (ASD) encompasses early onset neurological conditions that result in impairments in social behavior and communication with an estimated national prevalence of 1.7% [1].  While the actual cause of ASD remains unknown, the current understanding points towards this condition being a result of complex interactions between environmental and genetic factors. As environmental factors can be too numerous to look at in their entirety, composition analysis of urine collected from individuals diagnosed with ASD may provide some insight into the underlying nature of some of these factors and the metabolic pathways contributing to the disorder. Thus, the purpose of this investigation was to comprehensively assess the association between urinary element compositions and the presence of ASD in children and, separately, of mothers of children with ASD. It is important to point out that, while many studies have focused on the presence of toxins, the focus here is on analysis of essential elements which has received considerably less attention in the literature.

Materials and Methods: A cohort of 47 children, 21 with an ASD diagnosis and 26 who were developing typically (TD), and 59 mothers, 30 of whom had children with ASD and 29 who did not, were recruited. Children included were required to be between ages 2-5 and if they had an ASD diagnosis then this was verified via the ADI-R (Autism Diagnostic Interview-Revised). Exclusion criteria for mothers were the usage of supplements containing folic acid and/or vitamin B12 and if they were pregnant or were planning to become pregnant in the next six months. First morning urine samples were collected from the 106 participants and the samples were analyzed for 18 essential and 20 toxic elements by Doctor’s Data.  Statistical analysis was performed using a sample optimized univariate analysis for each element. However, no correction was performed for multiple hypothesis testing. Multivariate analysis using Fisher discriminant analysis (FDA) and support vector machine (SVM) were performed to ascertain the possibility of characterizing individuals as belonging to the ASD group based solely on their urinary element composition. Using elements that had been deemed significant (p-value <.05) by the univariate analysis, all permutations of 3-7 elements were assessed using FDA and separately by SVM. “Leave-one-out” cross-validation was used to ensure statistical independence of results.

Results and Discussion: Five essential elements were significantly lower in the ASD child group compared to the TD child group from the univariate analysis. Specifically, zinc (-23.4%, p-value=0.0438), selenium (-26.4%, p-value=0.0061), molybdenum (-33.5%, p-value=0.0053), phosphorous (-33.5%, p-value=0.0019), and sulfur (-35.8%, p-value=6.02E-05) levels were found to be lower in the child autism group. In contrast, chromium levels were found to be significantly greater in concentration within the autism child cohort (+20.8%, p-value=0.04). Additionally, different concentrations of the toxic metals mercury  (+553%, p-value=0.03 ), tungsten (-35.7%, p-value=0.0476) and tin (-48.05%,p-value=0.0030) were observed in the autism group of children  Among the mothers, there were no significant differences between groups, except for a slightly higher copper concentration in the urine of the ASD mothers (+9%, p-value=0.0347). The FDA model that was determined to best separate groups in the child cohort consisted of four elements, which were zinc, sulfur, tin, and phosphorus. For the same cohort, the SVM method determined a hyperplane that was capable of distinguishing 89% of samples accurately.  FDA and SVM analysis both revealed notable results for their performance on the child cohort, with 77% and 83% cross validation accuracy, respectively. It should be noted that the differences in essential elements between the two child cohorts are significantly larger than the differences of the toxic elements; this is important insofar as the focus in the existing literature has been more on toxic than on essential elements.

Conclusions:  Although there have been other studies which have sought to analyze the concentration of certain essential elements in urine between children with ASD and those that are TD, this study represents one of the most comprehensive analyses performed in terms of the breadth of elements examined. It was shown that there are statistically significant differences between the ASD and TD children cohort’s concentrations for nine specific elements. Notably, the concentrations of five essential elements in the urine of the children with ASD were lower than for their typically developing peers. This is particularly noteworthy as 46% of the children in the ASD cohort and only 24% of the TD cohort were taking multivitamins/minerals. The univariate analysis performed on the mother cohort revealed little significant difference between those that had children with ASD and those that did not.  Lastly, the use of multivariate techniques such as FDA and SVM further illustrated the differences between the element concentrations in the urine of the children in the ASD and TD groups. 

References: [1] J. Baio, et al,   MMWR Morb Mortal Wkly Rep, 2018 Nov 16, 67(45): 1279. Available:  10.15585/mmwr.mm6745a7