Artificial intelligence analysis of newborn leucocyte epigenomic markers for the prediction of autism

Abstract

Epigenetic modification appears to be important in the pathogenesis of Autism and Autism spectrum disorder (ASD). Early detection and intervention is known to correlate with improved long-term outcomes. There is thus intense scientific interest in the disease pathogenesis and early prediction. Our goal was to investigate the epigenetic basis of classic autism and identify early biomarkers. Cytosine (‘CpG’) methylation was measured genome-wide in 14 autism cases and 10 controlsusing the Infinium HumanMethylation450 BeadChip assay. Six Machine Learning/Artificial Intelligence (AI) platforms including Deep-Learning (DL) were then used for autism prediction. Ingenuity Pathway Analysis (IPA) was further used to interrogate autism pathogenesis by identifying over-represented biological pathways. Highly significant CpG methylation changes were found in 230 loci (249 genes). Using only individual markers meeting a stringent p-value threshold criterion (p < 5 x 10-8), achieved an AUC (95% CI) = 1.0 (0.80–1.00) with 97.5% sensitivity and 100.0% specificity. Conventional regression analysis using 3 CpG markers yielded an AUC (95% CI) = 1.00 (1.00–1.00) with 100.0% sensitivity and specificity each. Epigenetic dysregulation was identified in several important candidate genes including some previously linked to autism development e.g.: EIF4E, FYN, SHANK1 and VIM. We found epigenetic dysregulation of pathways involved in neuroinflammation signaling, synaptic long term potentiation, serotonin degradation, mTOR signaling and signaling by Rho-Family GTPases. Blood epigenomics and AI techniques achieved accurate newborn prediction of autism. Our results provide further evidence in support of a significant role of epigenetic alteration in autism pathogenesis.