mixMC

mixMC is a multivariate framework implemented in mixOmics for microbiome data analysis. The framework takes into account the inherent characteristics of microbiome data, namely sparsity (a large number of zeros in the data) and compositionality (occurring naturally in ecosystems, as well as resulting sequencing artefacts). The mixMC framework aims to identify key microbial communities associated with their habitat or environment.

mixMC addresses the limitations of existing multivariate methods for microbiome studies and proposes unique analytical capabilities: it handles compositional and sparse data, repeated-measures experiments and multiclass problems. It also highlights important discriminative features, and provides interpretable graphical outputs to better understand the microbial communities' contribution to each habitat. The framework from our paper is summarised below:

plot of chunk unnamed-chunk-1

Data

In the tabs under mixMC, examples are provided applying mixMC to microbiome 16S data sets, similar analyses would apply for whole genome shotgun sequencing data. The data are directly available through the mixOmics package. If you would like to download the full data sets and the associated R scripts used for the paper, then click on the following links:

Non-Repeated Measures analysis with the Koren data set

Repeated Measures analysis with the HMP most diverse body sites

Repeated Measures with the HMP Oral body sites 16S

How does mixMC fit into mixOmics?

mixMC is a pipeline set up for microbial communities, using some of the standards methods in mixOmics but with a bit of tweaking. The method sPLS-DA has been improved with CLR transformation and includes a multilevel decomposition for repeated measurements design that are commonly encountered in microbiome studies. The multilevel approach we developed in [4] enables the detection of subtle differences when high inter-subject variability is present due to microbial sampling performed repeatedly on the same subjects but in multiple habitats. To account for subject variability, the data variance is decomposed into within variation (due to habitat) and between subject variation [5], similar to a within-subjects ANOVA in univariate analyses.

As part of the pipeline, we added the new graphical output plotLoadings to visualise the OTUs selected by sPLS-DA on each component and the sample group or habitat in which the OTU is most (or least) abundant. See examples in our other tabs.

What is next?

  • In collaboration with colleagues from INRA Toulouse, France, the package mixKernel is available on our website (calls mixOmics) to integrate different types of data using kernel models (see here).

  • We are working on how to manage batch effects in microbiome studies, see [6] and soon a new multivariate method to correct for bacth effects.

References

  1. Lê Cao KA, Costello ME, Lakis VA, Bartolo F, Chua XY, et al. (2016) MixMC: A Multivariate Statistical Framework to Gain Insight into Microbial Communities. PLOS ONE 11(8): e0160169

  2. Aitchison, J., 1982. The statistical analysis of compositional data. Journal of the Royal Statistical Society. Series B (Methodological), pp.139-177.

  3. Filzmoser, P., Hron, K. and Reimann, C., 2009. Principal component analysis for compositional data with outliers. Environmetrics, 20(6), pp.621-632.

  4. Liquet, B., Lê Cao, K.A., Hocini, H. and Thiébaut, R., 2012. A novel approach for biomarker selection and the integration of repeated measures experiments from two assays. BMC bioinformatics, 13(1), p.325.

  5. Westerhuis, J.A., van Velzen, E.J., Hoefsloot, H.C. and Smilde, A.K., 2010. Multivariate paired data analysis: multilevel PLSDA versus OPLSDA. Metabolomics, 6(1), pp.119-128.

  6. Wang Y and Lê Cao K-A (2019). Managing Batch Effects in Microbiome Data. Briefings in Bioinformatics