Machine Learning Pipeline to Link Gut Microbiome Signature with Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition with growing evidence implicating the microbiota-gut-brain axis in its pathophysiology, highlighting the gut microbiome as a potential source of diagnostic biomarkers and therapeutic targets. Existing research is limited by cohort heterogeneity, inconsistent pre-processing pipelines, and poor cross-study reproducibility, which restrict the generalizability of reported microbial signatures. To address these gaps, the objective of this work is to develop a robust and interpretable machine learning pipeline. The pipeline is capable of identifying consistent gut microbiome signatures associated with ASD across independent datasets. This work integrates two publicly available cohorts GSE113690 (16S rRNA sequencing) and PRJNA516054 (shotgun metagenomics) harmonizes microbial features at the genus level. The harmonized features are applied for systematic pre-processing, feature extraction, and Random Forest classification, including permutation importance and SHAP analysis. The proposed model achieved an overall classification accuracy of 81%, demonstrating strong recall for ASD samples and identifying biologically relevant genera that consistently contributed to model predictions across datasets. These findings indicate that machine learning-driven integration of heterogeneous microbiome datasets can yield reproducible and interpretable microbial signatures, supporting the potential of gut microbiome profiling as a complementary tool for ASD diagnosis and advancing precision-based microbiome research in neurodevelopmental disorders.