The gut–brain axis (GBA) represents a complex, bidirectional communication network between the gut microbiome and the central nervous system, influencing both neurological health and the pathogenesis of various diseases. This review explores the integrative role of neuroimaging and machine learning (ML) in advancing our understanding of microbiota–brain interactions, emphasizing their combined potential to uncover novel biomarkers and therapeutic targets. Neuroimaging techniques, including functional magnetic resonance imaging (MRI), diffusion tensor imaging, and structural MRI, have revealed how gut microbiota imbalances, or dysbiosis, affect key brain networks and structural connectivity, contributing to cognitive dysfunction, emotional disturbances, and neurodegenerative conditions. ML methodologies, including deep learning (DL) and multimodal data fusion, are proving indispensable in extracting meaningful insights from high-dimensional neuroimaging and microbiome datasets. Supervised approaches, such as random forests and deep neural networks, have achieved high accuracy in predicting neurological outcomes based on microbial signatures, while unsupervised learning identifies distinct microbiota–brain connectivity patterns associated with disorders such as autism spectrum disorder and depression. Additionally, explainable AI (XAI) techniques are being increasingly applied to enhance the interpretability of ML-driven biomarker discovery, shedding light on the neuroprotective effects of butyrate-producing bacteria (e.g., Faecalibacterium , Roseburia ) and the potential for neuroinflammation linked to an overabundance of Proteobacteria . These findings point to the transformative potential of combining neuroimaging and ML in precision medicine, offering a new paradigm for the diagnosis and treatment of neurological disorders ranging from irritable bowel syndrome to Alzheimer’s disease. However, challenges related to data harmonization, generalizability across populations, and establishing causal relationships remain, necessitating further research to realize the full clinical potential of this approach.

MOTIVATION: Regulatory elements (REs), such as enhancers and promoters, are known as regulatory sequences functional in a heterogeneous regulatory network to control gene expression by recruiting transcription regulators and carrying genetic variants in a context specific way. Annotating those REs relies on costly and labor-intensive next-generation sequencing and RNA-guided editing technologies in many cellular contexts. RESULTS: We propose a systematic Gene Ontology Annotation method for Regulatory Elements (RE-GOA) by leveraging the powerful word embedding in natural language processing. We first assemble a heterogeneous network by integrating context specific regulations, protein-protein interactions and gene ontology (GO) terms. Then we perform network embedding and associate regulatory elements with GO terms by assessing their similarity in a low dimensional vector space. With three applications, we show that RE-GOA outperforms existing methods in annotating TFs' binding sites from ChIP-seq data, in functional enrichment analysis of differentially accessible peaks from ATAC-seq data, and in revealing genetic correlation among phenotypes from their GWAS summary statistics data. AVAILABILITY AND IMPLEMENTATION: The source code and the systematic RE annotation for human and mouse are available at https://github.com/AMSSwanglab/RE-GOA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.