Fig. 7
From: Modeling microbiome-trait associations with taxonomy-adaptive neural networks
MIOSTONE discovers microbiome-disease associations across different taxonomic levels. Feature attribution methods are used to interpret the MIOSTONE model and quantify the relationships between microbiome taxa and disease traits. a Three mainstream feature attribution methods—DeepLIFT, Integrated Gradients, and SHAP—demonstrate strong consistency in quantifying key microbiome-disease associations derived from the MIOSTONE model. b Feature importance attribution by DeepLIFT is performed at the genus, family, order, and class taxonomic levels. When emphasizing an important taxonomic group, the taxonomic subtree rooted at that group-along with its group-specific taxa-is highlighted for improved visualization. The top-ranked taxonomic groups are displayed with their respective names, supported by relevant literature evidence and accompanying PubMed identifiers