Autism spectrum disorder (ASD) is a neuro developmental disorder characterized by a series of abnormal social behaviors. The increasing prevalence of ASD has led to the discovery of a correlation with the intestinal microbiome in many studies. In our research, we evaluated 297 subjects, including 169 individuals with ASD and 128 neurotypical subjects, from the Sequence Read Archive database. We conducted a series of analyses, including alpha-diversity, phylogenetic profiles, and functional profiles, to explore the correlation between the gut microbiome and ASD. The principal component analysis (PCA) indicated that ASD and neurotypical subjects could be divided based on the unweighted UniFrac distance. The genera Prevotella, Roseburia, Ruminococcus, Megasphaera, and Catenibacterium might be biomarkers of ASD after linear discriminant analysis effect size (LEfSe) evaluation and Random Forest analysis, respectively. The functional analysis found six significant pathways between ASD and neurotypical subjects, including oxidative phosphorylation, nucleotide excision repair, peptidoglycan biosynthesis, photosynthesis, photosynthesis proteins, and two-component system. Based on these alterations of the intestinal microbiome in ASD subjects, we developed four machine learning models: random forest (RF), Multilayer Perceptron (MLP), kernelized support vector machines with the RBF kernel (SVMs), and Decision trees (DT). Notably, the RF model after RF selection was superior, with an F1 score of 0.74 and area under the curve of 0.827(0.004), suggesting the reliability and generalizability of predictive model. Besides, the validation performance of RF model after RF selection could be 0.75(0.01) on external cohort collected by our laboratory. Our study advances the understanding of human gut microbiome in ASD that designing and evaluating microbially based interventions of ASD.
Keywords: Austims spectrum disorders; Gut microbiome; Machine learning; Random forest.
Copyright © 2020 Elsevier Ltd. All rights reserved.