Getting-Started.Rmd

---
title: "Getting started with MuData for MultiAssayExperiment"
author:
- name: "Danila Bredikhin"
  affiliation: "European Molecular Biology Laboratory, Heidelberg, Germany"
  email: "danila.bredikhin@embl.de"
- name: "Ilia Kats"
  affiliation: "German Cancer Research Center, Heidelberg, Germany"
  email: "i.kats@dkfz-heidelberg.de"
date: "`r Sys.Date()`"
output:
  BiocStyle::html_document:
    toc_float: true
vignette: >
  %\VignetteIndexEntry{Getting started with MuDataMae}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
```

## Introduction

Multimodal data format — [MuData](https://mudata.readthedocs.io/) — 
[has been introduced](https://www.biorxiv.org/content/10.1101/2021.06.01.445670v1) 
to address the need for cross-platform standard for sharing 
large-scale multimodal omics data. Importantly, it develops ideas of and is 
compatible with [AnnData](https://anndata.readthedocs.io/) standard for storing 
raw and derived data for unimodal datasets.

In Bioconductor, multimodal datasets have been stored in 
[MultiAssayExperiment](https://bioconductor.org/packages/MultiAssayExperiment) 
(MAE) objects. This `MuData` package provides functionality to read data from 
MuData files into MAE objects as well as to save MAE objects into H5MU files.

## Installation

The most recent dev build can be installed from GitHub:

```{r, eval = FALSE}
library(remotes)
remotes::install_github("ilia-kats/MuData")
```

Stable version of `MuData` will be available in future bioconductor versions.

## Loading libraries

```{r setup, message = FALSE}
library(MuData)
library(MultiAssayExperiment)

library(rhdf5)
```

## Writing H5MU files

We'll use a simple MAE object from the `MultiAssayExperiment` package that 
we'll then save in a H5MU file.

```{r}
data(miniACC)
miniACC
```

We will now write its contents into an H5MU file with `WriteH5MU`:

```{r}
writeH5MU(miniACC, "miniacc.h5mu")
```

## Reading H5MU files

We can manually check the top level of the structure of the file:

```{r}
rhdf5::h5ls("miniacc.h5mu", recursive = FALSE)
```

Or dig deeper into the file:

```{r}
h5 <- rhdf5::H5Fopen("miniacc.h5mu")
h5&'mod'
rhdf5::H5close()
```

### Creating MAE objects from H5MU files

This package provides `ReadH5MU` to create an object with data from an H5MU 
file. Since H5MU structure has been designed to accommodate more structured 
information than MAE, only some data will be read. For instance, MAE has no 
support for loading multimodal embeddings or pairwise graphs.

```{r}
acc <- readH5MU("miniacc.h5mu")
acc
```

Importantly, we recover the information from the original MAE object:

```{r}
head(colData(miniACC)[,1:4])
head(colData(acc)[,1:4])
```

Features metadata is also recovered:

```{r}
head(rowData(miniACC[["gistict"]]))
head(rowData(acc[["gistict"]]))
```

#### Backed objects

It is possible to read H5MU files while keeping matrices (both `.X` and 
`.layers`) on disk.

```{r}
acc_b <- readH5MU("miniacc.h5mu", backed = TRUE)
assay(acc_b, "RNASeq2GeneNorm")[1:5,1:3]
```

The data in the assay is a `DelayedMatrix` object:

```{r}
class(assay(acc_b, "RNASeq2GeneNorm"))
```

This is in contrast to the `acc` object that has matrices in memory:

```{r}
assay(acc, "RNASeq2GeneNorm")[1:5,1:3]
class(assay(acc, "RNASeq2GeneNorm"))
```

## References

- [Muon: multimodal omics analysis framework](https://www.biorxiv.org/content/10.1101/2021.06.01.445670) preprint

- [mudata](https://mudata.readthedocs.io/) (Python) documentation

- muon [documentation](https://muon.readthedocs.io/) and [web page](https://gtca.github.io/muon/)

## Session Info

```{r}
sessionInfo()
```