Background: 
Autism spectrum disorder (ASD) is a genetically heterogeneous disorder and may becaused by both inherited and de novo gene variants. Increasing evidence points to acontribution of environmental and epigenetic factors in ASD, but their connectionsare still largely unexplored. 

Objectives: 
Aim of the present pilot study was to apply the Auto Contractive Map algorithm(Auto CM), a special kind of Artificial Neural Networks to link various ASD maternalrisk factors to the DNA methylation levels of selected genes, sex, and symptomsseverity (ADOS 2 score) of their children diagnosed with an ASD. 

Methods: 
A total of 58 ASD children aged less than 8 years (mean age 4.35 ± 1.79 years) wererecruited, including 23 males and 35 females. Blood DNA methylation levels of MECP2, BDNF, OXTR, RELN, BCL 2, EN 2, and HTR 1 A, were measured by means of MSHRM technique. Mothers filled in a detailed questionnaire on various environmentalfactors during pregnancy. We also investigated the methylation levels of miRNAencoding genes ( miR-30e, miR-23/27a, miR 28 miR 92 a 1 miR 92 a 2 miR 21) in DNAextracted from saliva of 11 idiopathic ASD and 13 typically developing preschoolaged girls. To graphically show the most important connections among variables weused a four generation artificial neural network called Auto-CM, that developsweights that are proportional to the strength of the associations of all variables eachother. The weights are then transformed in physical distances so that couples ofvariables whose connection weights are higher become nearer and vice versa. Afterthe training phase, the weights matrix of the Auto-CM represents the warpedlandscape of the dataset. Subsequently, a minimum spanning tree filter was appliedto the weights matrix of the Auto-CM system to obtain a map of the mainconnections between the variables of the dataset and the basic semantic of theirsimilarities, defined connectivity map (figure 1). 

Results: 
Sex differences were observed in blood DNA methylation levels of the studiedgenes, and ANNs revealed sex-specific connections among maternal risk factors andgene methylation. Furthermore, ANNs selected a set of variables allowingdiscriminating between high and low-moderate ADOS-2 scores with 86.8% overallaccuracy. Particularly, high gestational weight gain, lack of folic acid supplements,advanced maternal age, pre-term birth, low birthweight, and living in rural contextwere the best predictors of high ADOS-2 score. Moreover, the analysis of saliva DNAsamples revealed that Mir-28 methylation levels could represent a biomarker ofdisease severity in ASD children. 

Conclusions: 
ANNs revealed links among ASD maternal risk factors, symptoms severity and genemethylation levels, as well as sex differences in gene methylation levels that warrantfurther investigation in ASD.

Notes:

1 - University of Pisa, Italy

2 - IRCCS Fondazione Stella Maris, Calambrone (Pisa), Italy

3 - Villa Santa Maria Foundation Autism Research Unit, Tavernerio (Como), Italy