Summary and Setup

This lesson teaches how to analyse microbiome data together with other data types, such as metabolites and host biomarkers, measured from the same samples. Using data from a diet experiment in rats, you will learn to store several data tables in one object, to test whether they are related, to find the shared patterns in them, and to predict an outcome from all of them.

The lesson assumes that you can already analyse a single microbiome table, for example from the microbiome lesson (see the prerequisites below). It uses the OMA framework, a family of R/Bioconductor packages for microbiome data science, together with other Bioconductor packages for multi-omics. After the lesson, you can continue learning with the online book Orchestrating Microbiome Analysis (OMA).

Learning objectives

After this lesson, learners will be able to:

Learning objective Episode
Explain what multi-omics data is, why the microbiome is studied together with other data types, where the data comes from, and which questions the main groups of integration methods answer. Introduction
Store, explore, subset and prepare several data types of the same samples in a MultiAssayExperiment. MultiAssayExperiment and the study data
Test whether two data types are related, and find the features that link them, with the Mantel test, MiRKAT and pairwise associations. Exploratory methods
Find the main patterns that several data types share with joint RPCA and MOFA. Unsupervised methods
Predict an outcome from several data types and find the features that predict it with IntegratedLearner and DIABLO. Supervised methods

Where does this lesson fit?

A microbiome study goes from collecting samples and sequencing them, through processing the reads into tables, to analysing the tables. The analysis of one table covers the data object, data processing and statistical analysis. This lesson covers the next step: data integration, where the microbiome is combined with other data types.

Microbiome data science workflow with the data integration step highlighted: taxonomic profiles and metabolic profiles are combined in a joint analysis.
Microbiome data science workflow. This lesson covers the data integration step. Source: OMA.

The data

We analyse samples from 40 rats that were fed one of four diets: a high-fat or a low-fat diet, with or without a prebiotic supplement called xylo-oligosaccharides (XOS) (Hintikka et al. 2021). For every rat, the data contains three tables: the gut microbiota (16S rRNA sequencing), metabolites (NMR spectroscopy) and host biomarkers. The data comes with the mia package, so there is nothing to download.

Prerequisite

Prerequisites

  • The microbiome lesson, or equivalent experience: the TreeSummarizedExperiment class, agglomeration and transformation, beta diversity and ordination, and the basics of machine learning.
  • Basic R programming: data frames, factors, functions and simple for loops.
  • Basic statistics: what a p-value and a correlation are.

Software

This lesson uses R and RStudio, together with a few R/Bioconductor packages.

  1. Install R 4.6 or newer from CRAN.
  2. Install RStudio Desktop from Posit.

Packages

We install the packages ourselves, step by step, so that you see where each of them comes from. Run the commands below in the R console. The installation can take several minutes. If R asks whether to update other packages, you can answer a (all).

Package installers

Bioconductor packages are installed with BiocManager, and packages that are not yet in Bioconductor are installed from GitHub with remotes. Both come from CRAN:

R 

install.packages(c("BiocManager", "remotes"))

The lesson needs the current Bioconductor release. Check your version with:

R 

BiocManager::version()

Bioconductor and CRAN packages

BiocManager::install() installs packages from both Bioconductor and CRAN:

R 

BiocManager::install(c(
    "mia",                  # microbiome analysis and the example data
    "miaViz",               # microbiome visualization
    "MultiAssayExperiment", # container for multi-omics data
    "scater",               # PCA
    "UpSetR",               # UpSet plots of the samples
    "MOFA2",                # multi-omics factor analysis
    "mixOmics",             # DIABLO
    "randomForest",         # random forest, used by IntegratedLearner
    "ggplot2"               # plots
))

MOFA2 uses a few more packages for some of its plots and tests:

R 

BiocManager::install(c("psych", "corrplot", "pheatmap"))

GitHub packages

The multiomics and IntegratedLearner packages are not yet in Bioconductor, so we install them from GitHub:

R 

remotes::install_github("microbiome/multiomics")
remotes::install_github("himelmallick/IntegratedLearner")

Check the installation

Load each package. If a package was not installed, library() gives an error that names it:

R 

library(mia)
library(miaViz)
library(MultiAssayExperiment)
library(scater)
library(UpSetR)
library(MOFA2)
library(mixOmics)
library(multiomics)
library(IntegratedLearner)

Set up MOFA2

MOFA2 trains its models in Python. The first time you train a model, it installs its own Python environment. You do not need to install Python yourself, but this step can take 10 minutes or more, so please do it before the workshop. Run this code, which trains a small model on example data from MOFA2:

R 

library(MOFA2)
model <- create_mofa(make_example_data()$data)
model <- prepare_mofa(model)
model <- run_mofa(model, outfile = tempfile(fileext = ".hdf5"),
                  use_basilisk = TRUE)

When it prints Training finished, you are ready.

Callout

Problems with the installation?

If a package fails to install, check that your R version is recent enough with R.version.string, and try the installation again. If the problem persists, ask for help before the workshop, so that we can solve it together.

Project

Create a new RStudio project for the lesson (File → New Project → New Directory → New Project). The data comes with the mia package, so there is nothing to download.