Autism Spectrum Disorder (ASD) is a neurological, developmental disorder that affects communication and behaviour. Currently, there is a lack of understanding about the aetiology of ASD and children are not diagnosed reliably until they are at least 3 to 4 years of age. Henceforth, there is an urgent need for the identification of diagnostic biomarkers for the prediction of ASD.
Metabolomic profiling was undertaken on cord blood plasma from ASD cases vs gender matched controls from the Cork BASELINE Birth Cohort using Liquid chromatography-tandem mass spectrometry.
Clinical data (maternal and child BMI, birth weight, gestational age, health and lifestyle etc.) were combined with metabolite peaks to explore machine learning (ML) algorithms: t-test, partial least squares-determinant analysis (PLS-DA) and random forest (RF).
We profiled 2612 compounds and 18 clinical variables to identify the top predictive metabolite and clinical features for the development of ASD. RF feature selection resulted in the highest performing 20 metabolite subset, and pathway analysis revealed significant alterations in 5 metabolic pathways.
Dr Kirsten Dowling will deliver her presentation on Friday, February 11 at 12 pm.