Complex network and artificial intelligence combined approach to investigate autism spectrum disorder through gene expression data.

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder where only 20% of cases can be explained by known genetic mutations. Recent studies have identified a relationship between gene expressions and ASD through brain transcriptome analysis. In our study we analyzed a publicly available dataset of gene expression from 94 individuals (51 healthy and 43 autistic) to identify gene communities most related to ASD. Through a data-driven approach based on complex networks, we have grouped thousands of genes into small stable communities by exploiting the spin-glass algorithm inspired by the model of Nobel Prize winner Giorgio Parisi. The Boruta algorithm, a wrapper method, was then applied to select the most discriminating genes within each community. Finally, Random Forest (RF), a machine learning algorithm, was used to classify the 29 communities found via complex network approach; RF provides a classification accuracy ranging from 70.86% to 77.61%. Our results suggest that using gene expression and artificial intelligence may help identify new ASD-related biomarkers, but further studies are needed to validate these gene communities biologically and statistically.