| name | bioconductor-dirichletmultinomial |
| description | Dirichlet-multinomial mixture models can be used to describe variability in microbial metagenomic data. This package is an interface to code originally made available by Holmes, Harris, and Quince, 2012, PLoS ONE 7(2): 1-15, as discussed fu |
| when_to_use | Use when: Fitting Dirichlet-Multinomial models to microbial metagenomic count data using dmn().; Identifying the optimal number of Dirichlet components (clusters) by comparing laplace(), AIC(), or BIC() scores.; Building generative classifiers for phenotypic groups using dmngroup().; Assigning new samples to classes using predict().. Not for: For continuous covariate modeling, use alternative regression frameworks because dmngroup() is designed for discrete phenotypic categories (e.g., 'Lean', 'Obese'). |
| user-invocable | false |
DirichletMultinomial
Dependencies & Environment
Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.
- Version: 1.54.0 · Bioconductor: 3.23 · R: ≥ 4.6
- Depends: S4Vectors, IRanges
- Imports: BiocGenerics
- System requirements: gsl
- Install:
BiocManager::install("DirichletMultinomial")
When to Use
- Fitting Dirichlet-Multinomial models to microbial metagenomic count data using
dmn().
- Identifying the optimal number of Dirichlet components (clusters) by comparing
laplace(), AIC(), or BIC() scores.
- Building generative classifiers for phenotypic groups using
dmngroup().
- Assigning new samples to classes using
predict().
When NOT to Use
- For continuous covariate modeling, use alternative regression frameworks because
dmngroup() is designed for discrete phenotypic categories (e.g., 'Lean', 'Obese').
Data Requirements
- A matrix of raw integer counts (samples in rows, taxa in columns).
- Phenotypic information as a factor for classification.
Key Parameters
- count (default): The matrix of samples by taxa counts provided to
dmn() or dmngroup().
- k (default): The number of Dirichlet components to model in
dmn() or dmngroup().
- assign (FALSE): Set to
TRUE in predict() or mixture() to return the component/class with the maximum value.
- scale (FALSE): Set to
TRUE in fitted() to scale the posterior mean difference by theta.
Best Practices
- Fit models across a range of components (e.g.,
k=1:7) and select the best fit using which.min(lplc) on the Laplace scores.
- Use
cvdmngroup() to perform cross-validation and assess classifier performance via ROC curves.
- Use
heatmapdmn() to visualize samples arranged by their dominant Dirichlet component.
Common Pitfalls
- Cross-validation with
cvdmngroup() is computationally expensive; fix this by using mc.preschedule=FALSE and running it on multiple cores via the parallel package.
- Passing non-integer or normalized data to
dmn() violates the multinomial assumption; fix this by ensuring the input is a raw count matrix.
Alternatives
vegan: For standard distance-based clustering and ordination rather than generative mixture modeling.phyloseq: For general microbiome data manipulation and visualization.
Citations
References
Run this on BioMate
This skill is the knowledge layer — when, why, and how to use dirichletmultinomial. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.
▶ Open dirichletmultinomial on BioMate →
