An Exponential Scale Mixture Model for Metatranscriptomics Data with Application to Inflammatory Bowel Disease

Metatranscriptomic (MTX) sequencing enables profiling of gene expression across microbial communities, providing a framework for linking genetic potential with functional activity. However, standard pipelines report normalized abundances rather than raw counts, limiting the use of count-based RNA-seq methods, while Gaussian-based alternatives rely on transformations and assumptions that are often poorly suited to MTX data. We propose a new modeling framework for differential expression analysis of MTX data, built on a scale mixture of exponential distributions, that incorporates DNA abundance to adjust for genomic potential, accommodates subject-specific random effects, treats zeros as left-censored, and employs a mixture prior to handle extreme sparsity. Applied to the IBDMDB multi-omics cohort, differential expression results vary substantially across models, including among Gaussian approaches with different pseudocount choices. Our approach identifies a distinct subset of candidate genes not detected by existing Gaussian methods; these may provide useful leads toward a novel understanding of transcriptomic patterns associated with dysbiosis in inflammatory bowel disease. Estimated dysbiosis effect directions are consistent between our model and Gaussian-based approaches, while effect sizes from our model tend to be larger in absolute value.