Table 8.

Summary of recent and representative studies aiming to distinguish individuals with ASD from TD individuals using multivariate analysis of patterns in gene expression and epigenetic activity. 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
Gene Expression
Glatt et al. (2012)218	60 infants and toddlers at risk for ASD and 68 TD controls	Evaluated children’s profiles of messenger RNA expression in peripheral blood mononuclear cells	Expression intensities of 48 probes	Radial basis function SVM	Predicted a replication sample (half of samples) with 93% sensitivity, 88% specificity, and 0.91 AUROC
Kong et al. (2012)219	66 (104) children with ASD and 33 (82) non-ASD controls for training (validation)	Profiling of blood gene expression levels in participants	55 genes	Partial least squares	Obtained 0.98 AUROC in the training set and 0.70 AUROC (68% accuracy) in the validation set
Hu and Lai (2013)220	87 individuals with ASD and 29 non-ASD individuals	Gene expression profiling of lymphoblastoid cell lines using DNA microarrays	74 genes	SVM	Achieved 91% sensitivity and 61% specificity with leave-one-out cross-validation
Latkowski and Osowski (2015)221	82 children with ASD and 64 TD children	Used gene expression data from a publicly available database	Unspecified number of genes used in ensemble classifier	Gaussian kernel SVM with ensemble of classifiers	Classified with 96% sensitivity and 83% specificity with ten-fold cross-validation
Pramparo et al. (2015)222	87 (44) toddlers with ASD and 55 (29) non-ASD toddlers for discovery (replication)	Profiling of leukocyte RNA expression in participants	Four co-expression modules containing 762 unique genes	Logistic regression	Achieved 75% accuracy, 77% sensitivity, and 72% specificity in replication set
Guan et al. (2016)223	104 children with ASD and 82 non-ASD controls	Used data on peripheral blood gene expression from Kong et al. (2012)	Three unique sets of five genes	Distance from multivariate centroid	In the validation set (half of samples), classified with 72%−76% accuracy
Nazeen et al. (2016)224	671 total samples from human ASD studies	Used high-throughput gene expression data from data repositories for conditions that co-occur with ASD	Genes overlapping the chemokine and Toll-like receptor signaling pathways	SVM, among others	Classified ASD versus non-ASD with average 70% classification accuracy with three-fold cross-validation
Oh et al. (2017)225	21 young adults with ASD and 21 TD controls	Used a microarray data set publicly available from a database	19 differentially expressed probes	SVM, k-nearest neighbors, discriminant analysis	Achieved up to 94% accuracy, 100% sensitivity, and 87.5% specificity on the validation set (16 samples)
Epigenetic Activity
Mundalil Vasu et al. (2014)226	55 individuals with ASD and 55 TD controls	Measured microRNA profiles in serum of participants	Five differentially expressed microRNAs	ROC analysis†	Classified with AUROC up to 0.91, with associated 85% sensitivity, 87% specificity
Hicks et al. (2016)227	24 children with ASD and 21 TD children	Measured salivary microRNA levels	14 top-ranked microRNAs	Partial least squares	Classified with 100% sensitivity and 96% specificity (AUROC = 0.97).
Cirnigliaro et al. (2017)228	30 children with ASD and 25 TD children	Profiled serum expression of microRNAs	One microRNA, miR-140–3p	Logistic regression	Averaged 63% sensitivity and 68% specificity with 100-random subsampling cross-validation
Hicks et al. (2018)142	238 children with ASD and 218 non-ASD children	Measured salivary levels of five subtypes of RNA, including microRNA	32 RNAs	Radial kernel SVM	Predicted the test set (84 total samples) with 82% sensitivity and 88% specificity (AUROC = 0.88)

^†Study performs classification, but only through univariate approaches.