Table 9.

Summary of recent and representative studies aiming to distinguish individuals with ASD from TD individuals using multivariate analysis of potential blood-based metabolite biomarkers. Reported sample sizes are the numbers used for classification and do not necessarily reflect the study’s total sample size.

Reference	Study Participants	Experimental Methods	Key Features	Multivariate Technique	Key Results
Momeni et al. (2012)229	22 children with ASD and 27 TD children	Analyzed plasma protein/peptide concentrations using mass spectrometry	Three differentially expressed peptides	Discriminant analysis	Classification of samples without hemolysis yielded 95% sensitivity and 85% specificity
West et al. (2014)230	52 children with ASD and 30 TD children	Measured concentrations of plasma metabolites through various mass spectrometry-based techniques	80 or 160 metabolites depending on classifier	SVM, partial least squares discriminant analysis	Predicted 21-sample validation set with AUROC of 0.84 (SVM) and 0.81 (partial least squares)
Wang et al. (2016)231	73 (100) children with ASD and 63 (100) TD children in discovery (validation) sets	Quantified serum metabolites with ultra-performance liquid chromatography and mass spectrometry	Docosahexaenoic acid and sphingosine 1-phosphate	Logistic regression	Achieved 90% sensitivity and 74% specificity for predicting the validation set
Howsmon et al. (2017)158	83 children with ASD and 76 TD children	Analyzed levels of plasma markers related to DNA methylation and oxidative stress	Seven transmethylation/transsulfuration measurements	Discriminant analysis	Classified with 98% sensitivity and 96% specificity using leave-one-out cross-validation
Anwar et al. (2018)232	38 children with ASD and 31 TD children	Investigated protein damage through quantification of glycation end-products in plasma and urine analysis	Four plasma protein adduct residues and two amino acids	SVM, among several other techniques	Observed 89% accuracy, 90% sensitivity, and 87% specificity using two-fold cross-validation
Barone et al. (2018)233	83 children with ASD and 79 TD children	Quantified acyl-carnitine and amino acid levels from dried blood spot specimens collected at time of the study	Eight acyl-carnitines	Naive Bayes	Predicted a 38-sample holdout set with 73% sensitivity and 63% specificity
Chen et al. (2018)234	32 children with ASD and 20 TD children	Profiled the serum proteome using fractionation and mass spectrometry techniques	Eight differentially expressed protein peaks	k-nearest neighbors	Achieved 99% sensitivity and 87% specificity using cross-validation
Shen et al. (2018)235	30 children with ASD and 30 TD children	Used isobaric tags for relative and absolute quantitation to measure medium- and low-abundance plasma proteins	Five plasma proteins	Combined ROC analysis	Classified with 0.98 AUROC, better than the AUROCs of the individual proteins
Howsmon et al. (2018)163	154 children and adolescents with ASD, compiled from three clinical trials	Validated classification with DNA methylation/oxidative stress markers presented by Howsmon et al. (2017)158	Five transmethylation/transsulfuration measurements	Discriminant analysis, among other techniques	Predicted an independent validation set of individuals with ASD with up to 88% sensitivity
Smith et al. (2019)236	253 (263) infants with ASD and 85 (79) TD infants in training (testing) set	Examined amino acid dysregulation metabotypes (AADMs) in blood plasma	Six AADMs	Ratios of AADMs to different amino acids	In the test set, sensitivities ranged from 8–14% and specificities ranged from 92100%
Zou et al. (2019)165	89 children with ASD and 89 TD children	Measured serum concentrations of folate-related metabolites	Six folate-related markers	Discriminant analysis	Correctly classified 84% of participants using leave-one-out cross-validation (87% sensitivity, 85% specificity)