This framework applies structural equation modeling (SEM) combined with linear mixed-effects models to analyze microbiome data.
It integrates penalized quasi-likelihood estimation with a debiased lasso for robust variable selection and inference.
Indirect effects are assessed using a non-parametric bootstrap procedure, with statistical significance evaluated through bias-corrected and accelerated (BCa) confidence intervals.
The current implementation demonstrates the approach on a 16S rRNA sputum microbiome dataset, but the framework can be adapted to other omics data.
Koldaş SS, Sezerman OU, Timuçin E. Exploring the role of microbiome in cystic fibrosis clinical outcomes through a mediation analysis. mSystems (2025).
https://doi.org/10.1128/msystems.00196-25