A robust microbiome signature for autism spectrum disorder across different studies using machine learning

Area of Science:

• Microbiome research
• Neurodevelopmental disorders
• Genomics and bioinformatics

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with varied comorbidities.
• Individuals with ASD frequently experience gastrointestinal issues and exhibit distinct gut microbial compositions.
• Existing gut microbiome studies in ASD lack consensus on specific bacterial taxa involved.

Purpose of the Study:

• To identify a core set of bacterial taxa associated with ASD classification using a sibling-controlled dataset.
• To validate these findings across independent cohorts, accounting for confounding factors like lifestyle.
• To explore the potential of machine learning for microbiome-based ASD prediction.

Main Methods:

• Applied recursive ensemble feature selection (REFS) to 16S rRNA gene sequencing data.
• Analyzed data from 117 subjects (60 ASD cases, 57 siblings) in the primary cohort.
• Validated findings using tenfold cross-validation on two independent cohorts (223 samples total).

Main Results:

• Identified 26 bacterial taxa that effectively discriminate ASD cases from controls in the sibling-controlled dataset.
• Achieved an average area under the curve (AUC) of 81.6% for ASD prediction in the primary cohort.
• Obtained average AUCs of 74.8% and 74% in the two independent validation cohorts.

Conclusions:

• The gut microbiome is strongly associated with ASD, suggesting it as a potential therapeutic target.
• The identified bacterial taxa can predict ASD status in children across multiple cohorts.
• The REFS approach can be utilized for identifying microbiome signatures in other 16S rRNA gene sequencing datasets.