mtradeR

mtradeR models metagenomic taxon trajectories to identify OTU-specific predictors of disease onset using a joint model with matching and regularization (JMR).

Key Features:

• Joint Modeling (JMR): Implements a joint model with matching and regularization (JMR) to link OTU longitudinal trajectories to host disease status.
• Nested Random Effects: Models between- and within-matched-set heterogeneity in OTU relative abundance and disease risk using nested random effects.
• Transformation-free Analysis: Operates on untransformed relative abundance data to preserve compositional structure.
• Regularization and Covariate Adjustment: Pre-selects top-correlated taxa using Bray-Curtis distance and elastic net regression and regularizes these longitudinal covariates to adjust for negative correlations in microbiota composition.
• Simulation Pipeline: Includes a simulation pipeline that generates true biomarkers while controlling pseudo-biomarkers caused by compositional effects.
• False Discovery and Pseudo-biomarker Control: Applies controls for false discovery rates and pseudo-biomarkers to enhance power for detecting true disease-associated microbial trajectories.

Scientific Applications:

• Infant fecal microbiome biomarker identification: Identifies taxa in infants' fecal samples whose pre-onset dynamics predict host disease status.
• Comparative evaluation and cohort analysis: Demonstrates superior performance on simulated datasets and real-world data such as the TEDDY cohort for detecting disease-associated microbial features.

Methodology:

mtradeR applies a joint model with matching and regularization (JMR) and nested random effects to untransformed OTU relative-abundance data, pre-selects taxa via Bray-Curtis distance and elastic net regression, and uses simulation together with false-discovery and pseudo-biomarker controls to detect OTU-specific trajectories.