Abstract
The gut microbiota plays a crucial role in human health, with growing evidence linking its composition to the development of Autism Spectrum Disorder. However, inconsistencies in previous studies have hindered the identification of a definitive microbial signature associated with Autism Spectrum Disorder. Machine learning models have emerged as powerful tools for analyzing microbiome data, yet their interpretability remains limited. In this study, we integrate in silico simulations with machine learning predictions to explore microbial interactions under different dietary conditions and provide biological context to features of the intestinal microbiota that are linked to Autism Spectrum Disorder. This study employs constraint-based modeling to simulate metabolic exchanges among key bacterial taxa in order to assess their ecological relationships. Findings reveal that high-fiber diets foster mutualistic and balanced interactions, whereas Western-style diets promote competitive and parasitic dynamics, potentially contributing to gut dysbiosis in Autism Spectrum Disorder. In addition, the presence of oxygen (a factor associated with colonocyte permeability, a pathological condition of the colon) significantly alters microbial interactions, influencing metabolic dependencies and the overall structure of the community. This integrative approach enhances the interpretability of machine learning-based Autism Spectrum Disorder classifiers, bridging computational predictions with mechanistic insights. By identifying diet-dependent microbial interactions, our study highlights potential dietary interventions to modulate the composition of the gut microbiota in Autism Spectrum Disorder. These findings underscore the value of combining in silico modeling and machine learning for unraveling complex microbiome-host relationships and improving Autism Spectrum Disorder biomarker identification.