AUTHOR=Tang Huitao , Liang Jiawei , Chai Keping , Gu Huaqian , Ye Weiping , Cao Panlong , Chen Shufang , Shen Daojiang 

TITLE=Artificial intelligence and bioinformatics analyze markers of children's transcriptional genome to predict autism spectrum disorder

JOURNAL=Frontiers in Neurology

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2023.1203375

DOI=10.3389/fneur.2023.1203375

ISSN=1664-2295

ABSTRACT=Autistic Spectrum Disorder (ASD) manifests as aberrations in social cognition and communication, circumscribed interests, and repetitive motor behaviors. Early diagnosis and intervention are seminal to mitigating sequelae, yet diagnosing ASD in toddlers remains recalcitrant.
In this study, we established a machine learning classification model based on biological features of ASD using mRNA expression data of peripheral blood from 128 ASD toddlers and 126 Control (CON) toddlers. We identified differentially expressed genes (DEGs) between ASD and CON, such as UBE4B, SPATA2, RBM3, etc. These genes are mainly distributed in immune-related, growth factor and neurotransmitter pathways. We screened 21 genes as key biomarkers using Least absolute shrinkage and selection operator (LASSO) regression and found that the Logistic regression accuracy was 0.86. Similarly, the constructed neural network based on the expression of these 21 genes also showed improved classification performance (AUC=0.88). Our findings suggest that the neurotransmitters and immune related 21 biomarkers identified by bioinformatics analysis and machine learning may be favorable for early diagnosis of ASD.