Plataforma não invasiva para detecção de Transtorno do Espectro Autista baseada em Lipidômica Salivar aliada à Inteligência Artificial

Abstract

Timely and accurate diagnosis of Autism Spectrum Disorder (ASD) in clinical settings remains a challenge due to the lack of diagnostic markers. The application of untargeted lipidomics has emerged as a powerful analysis for detecting novel biomarkers. In this context, the detection of salivary lipids may offer large-scale, non-invasive detection for ASD. In this study, we utilized untargeted lipidomics based on gas chromatography-mass spectrometry, coupled with artificial intelligence algorithms, to identify alterations in the salivary lipid profile as an alternative for ASD detection. We analyzed 33 saliva samples from children, including 18 neurotypical and 15 with a diagnosis of ASD who were not using medication. We identified 180 salivary lipids, with 27 being expressed with a frequency filter of 75%, and some salivary lipids were identified for the first time in saliva. Three salivary lipids—1-Naphthalenamine, N-phenyl-; hexadecanoic acid, 2- hydroxy-1-(hydroxymethyl)ethyl ester; and 13-Docosenamide, (Z)—had reduced expression (p<0.05) in ASD compared to neurotypical individuals. We also tested 6 artificial intelligence algorithms based on Python, and the Neural Network algorithm had the best performance, achieving 84% accuracy, 80% specificity, and 88% sensitivity, with the highest SHAP score for the lipid 13-Docosenamide. In summary, these data highlight the potential of salivary lipid profiling coupled with artificial intelligence algorithms as a non-invasive tool for screening children with ASD, enabling advances in precision medicine and personalized dentistry.