pbmc_multiomic.Rmd

---
title: "Joint RNA and ATAC analysis: 10x multiomic"
date: 'Compiled: `r format(Sys.Date(), "%B %d, %Y")`'
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

In this vignette, we'll demonstrate how to jointly analyze a single-cell dataset
measuring both DNA accessibility and gene expression in the same cells using
Signac and Seurat. In this vignette we'll be using a publicly available 10x
Genomic Multiome dataset for human PBMCs.

<details>
  <summary>**View data download code**</summary>

You can download the required data by running the following lines in a shell:

```{sh eval=FALSE}
wget https://cf.10xgenomics.com/samples/cell-arc/1.0.0/pbmc_granulocyte_sorted_10k/pbmc_granulocyte_sorted_10k_filtered_feature_bc_matrix.h5
wget https://cf.10xgenomics.com/samples/cell-arc/1.0.0/pbmc_granulocyte_sorted_10k/pbmc_granulocyte_sorted_10k_atac_fragments.tsv.gz
wget https://cf.10xgenomics.com/samples/cell-arc/1.0.0/pbmc_granulocyte_sorted_10k/pbmc_granulocyte_sorted_10k_atac_fragments.tsv.gz.tbi
```

</details>

```{r include=FALSE}
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

if (!requireNamespace("EnsDb.Hsapiens.v86", quietly = TRUE))
    BiocManager::install("EnsDb.Hsapiens.v86")

if (!requireNamespace("BSgenome.Hsapiens.UCSC.hg38", quietly = TRUE))
    BiocManager::install("BSgenome.Hsapiens.UCSC.hg38")

if (!requireNamespace("remotes", quietly = TRUE))
    install.packages("remotes")

if (!requireNamespace("SeuratDisk", quietly = TRUE))
    remotes::install_github("mojaveazure/seurat-disk")

if (!requireNamespace("glmGamPoi", quietly = TRUE))
    BiocManager::install("glmGamPoi")
```

```{r message=FALSE, warning=FALSE}
library(Signac)
library(Seurat)
library(EnsDb.Hsapiens.v86)
library(BSgenome.Hsapiens.UCSC.hg38)
```

```{r message=FALSE, warning=FALSE}
# load the RNA and ATAC data
counts <- Read10X_h5("../vignette_data/multiomic/pbmc_granulocyte_sorted_10k_filtered_feature_bc_matrix.h5")
fragpath <- "../vignette_data/multiomic/pbmc_granulocyte_sorted_10k_atac_fragments.tsv.gz"
```

```{r message=FALSE, warning=FALSE}
# get gene annotations for hg38
annotation <- GetGRangesFromEnsDb(ensdb = EnsDb.Hsapiens.v86)
seqlevels(annotation) <- paste0('chr', seqlevels(annotation))

# create a Seurat object containing the RNA adata
pbmc <- CreateSeuratObject(
  counts = counts$`Gene Expression`,
  assay = "RNA"
)

# create ATAC assay and add it to the object
pbmc[["ATAC"]] <- CreateChromatinAssay(
  counts = counts$Peaks,
  sep = c(":", "-"),
  fragments = fragpath,
  annotation = annotation
)
```

```{r}
pbmc
```

## Quality control

We can compute per-cell quality control metrics using the DNA accessibility data
and remove cells that are outliers for these metrics, as well as cells with
low or unusually high counts for either the RNA or ATAC assay.

```{r fig.width=18, message=FALSE, warning=FALSE}
DefaultAssay(pbmc) <- "ATAC"

pbmc <- NucleosomeSignal(pbmc)
pbmc <- TSSEnrichment(pbmc)
```

The relationship between variables stored in the object metadata can
be visualized using the `DensityScatter()` function. This can also be used
to quickly find suitable cutoff values for different QC metrics by setting
`quantiles=TRUE`:

```{r}
DensityScatter(pbmc, x = 'nCount_ATAC', y = 'TSS.enrichment', log_x = TRUE, quantiles = TRUE)
```

```{r}
VlnPlot(
  object = pbmc,
  features = c("nCount_RNA", "nCount_ATAC", "TSS.enrichment", "nucleosome_signal"),
  ncol = 4,
  pt.size = 0
)
```

```{r}
# filter out low quality cells
pbmc <- subset(
  x = pbmc,
  subset = nCount_ATAC < 100000 &
    nCount_RNA < 25000 &
    nCount_ATAC > 1800 &
    nCount_RNA > 1000 &
    nucleosome_signal < 2 &
    TSS.enrichment > 1
)
pbmc
```
<!-- ## Peak calling -->

<!-- The set of peaks identified using Cellranger often merges distinct peaks that are -->
<!-- close together. This can create a problem for certain analyses, particularly motif  -->
<!-- enrichment analysis and peak-to-gene linkage. To identify a more accurate set of peaks, -->
<!-- we can call peaks using MACS2 with the `CallPeaks()` function. Here we call peaks -->
<!-- on all cells together, but we could identify peaks for each group of cells separately -->
<!-- by setting the `group.by` parameter, and this can help identify peaks specific to -->
<!-- rare cell populations.  -->

<!-- ```{r include=FALSE} -->
<!-- old_path <- Sys.getenv("PATH") -->
<!-- Sys.setenv(PATH = paste(old_path, "/Users/stuartt/mambaforge/bin", sep = ":")) -->
<!-- ``` -->

<!-- ```{r message=FALSE, warning=FALSE} -->
<!-- # call peaks using MACS2 -->
<!-- peaks <- CallPeaks(pbmc) -->

<!-- # remove peaks on nonstandard chromosomes and in genomic blacklist regions -->
<!-- peaks <- keepStandardChromosomes(peaks, pruning.mode = "coarse") -->
<!-- peaks <- subsetByOverlaps(x = peaks, ranges = blacklist_hg38_unified, invert = TRUE) -->
<!-- ``` -->

<!-- ```{r include=FALSE} -->
<!-- gc() -->
<!-- ``` -->

<!-- ```{r message=FALSE, warning=FALSE} -->
<!-- # quantify counts in each peak -->
<!-- macs2_counts <- FeatureMatrix( -->
<!--   fragments = Fragments(pbmc), -->
<!--   features = peaks, -->
<!--   cells = colnames(pbmc) -->
<!-- ) -->
<!-- ``` -->

<!-- ```{r include=FALSE} -->
<!-- gc() -->
<!-- ``` -->

<!-- ```{r} -->
<!-- # create a new assay using the MACS2 peak set and add it to the Seurat object -->
<!-- pbmc[["peaks"]] <- CreateChromatinAssay( -->
<!--   counts = macs2_counts, -->
<!--   fragments = fragpath, -->
<!--   annotation = annotation -->
<!-- ) -->
<!-- ``` -->

## Gene expression data processing

We can normalize the gene expression data using SCTransform, and reduce the dimensionality
using PCA.

```{r results='hide', message=FALSE, warning=FALSE}
DefaultAssay(pbmc) <- "RNA"
pbmc <- SCTransform(pbmc)
pbmc <- RunPCA(pbmc)
```

## DNA accessibility data processing

Here we process the DNA accessibility assay the same way we would process a 
scATAC-seq dataset, by performing latent semantic indexing (LSI).

```{r message=FALSE, warning=FALSE}
DefaultAssay(pbmc) <- "ATAC"
pbmc <- FindTopFeatures(pbmc, min.cutoff = 5)
pbmc <- RunTFIDF(pbmc)
pbmc <- RunSVD(pbmc)
```

## Annotating cell types

To annotate cell types in the dataset we can transfer cell labels from an existing
PBMC reference dataset using tools in the Seurat package. See the
Seurat reference mapping [vignette](https://satijalab.org/seurat/v4.0/reference_mapping.html)
for more information.

We'll use an annotated PBMC reference dataset from [Hao et al. (2020)](https://www.biorxiv.org/content/10.1101/2020.10.12.335331v1),
available for download here: https://atlas.fredhutch.org/data/nygc/multimodal/pbmc_multimodal.h5seurat

Note that the SeuratDisk package is required to load the reference dataset.
Installation instructions for SeuratDisk can be found [here](https://github.com/mojaveazure/seurat-disk).

```{r message=FALSE, warning=FALSE}
library(SeuratDisk)

# load PBMC reference
reference <- LoadH5Seurat("../vignette_data/multiomic/pbmc_multimodal.h5seurat", assays = list("SCT" = "counts"), reductions = 'spca')
reference <- UpdateSeuratObject(reference)

DefaultAssay(pbmc) <- "SCT"

# transfer cell type labels from reference to query
transfer_anchors <- FindTransferAnchors(
  reference = reference,
  query = pbmc,
  normalization.method = "SCT",
  reference.reduction = "spca",
  recompute.residuals = FALSE,
  dims = 1:50
)

predictions <- TransferData(
  anchorset = transfer_anchors, 
  refdata = reference$celltype.l2,
  weight.reduction = pbmc[['pca']],
  dims = 1:50
)

pbmc <- AddMetaData(
  object = pbmc,
  metadata = predictions
)

# set the cell identities to the cell type predictions
Idents(pbmc) <- "predicted.id"

# remove low-quality predictions
pbmc <- pbmc[, pbmc$prediction.score.max > 0.5]
```

## Joint UMAP visualization

Using the weighted nearest neighbor methods in [Seurat v4](https://www.biorxiv.org/content/10.1101/2020.10.12.335331v1),
we can compute a joint neighbor graph that represent both the gene expression and 
DNA accessibility measurements.

```{r message=FALSE, warning=FALSE}
# build a joint neighbor graph using both assays
pbmc <- FindMultiModalNeighbors(
  object = pbmc,
  reduction.list = list("pca", "lsi"), 
  dims.list = list(1:50, 2:40),
  modality.weight.name = "RNA.weight",
  verbose = TRUE
)

# build a joint UMAP visualization
pbmc <- RunUMAP(
  object = pbmc,
  nn.name = "weighted.nn",
  assay = "RNA",
  verbose = TRUE
)

DimPlot(pbmc, label = TRUE, repel = TRUE, reduction = "umap") + NoLegend()
```

## Linking peaks to genes

For each gene, we can find the set of peaks that may regulate the gene by by
computing the correlation between gene expression and accessibility at nearby
peaks, and correcting for bias due to GC content, overall accessibility, and 
peak size. See the [Signac paper](https://www.biorxiv.org/content/10.1101/2020.11.09.373613v1)
for a full description of the method we use to link peaks to genes.

Running this step on the whole genome can be time consuming, so here we demonstrate
peak-gene links for a subset of genes as an example. The same function can be used
to find links for all genes by omitting the `genes.use` parameter:

```{r message=FALSE, warning=FALSE}
DefaultAssay(pbmc) <- "ATAC"

# first compute the GC content for each peak
pbmc <- RegionStats(pbmc, genome = BSgenome.Hsapiens.UCSC.hg38)

# link peaks to genes
pbmc <- LinkPeaks(
  object = pbmc,
  peak.assay = "ATAC",
  expression.assay = "SCT",
  genes.use = c("LYZ", "MS4A1")
)
```

We can visualize these links using the `CoveragePlot()` function, or alternatively
we could use the `CoverageBrowser()` function in an interactive analysis:

```{r fig.height=10, message=FALSE, warning=FALSE}
idents.plot <- c("B naive", "B intermediate", "B memory",
                 "CD14 Mono", "CD16 Mono", "CD8 TEM", "CD8 Naive")

p1 <- CoveragePlot(
  object = pbmc,
  region = "MS4A1",
  features = "MS4A1",
  expression.assay = "SCT",
  idents = idents.plot,
  extend.upstream = 500,
  extend.downstream = 10000
)

p2 <- CoveragePlot(
  object = pbmc,
  region = "LYZ",
  features = "LYZ",
  expression.assay = "SCT",
  idents = idents.plot,
  extend.upstream = 8000,
  extend.downstream = 5000
)

patchwork::wrap_plots(p1, p2, ncol = 1)
```

```{r include=FALSE}
saveRDS(object = pbmc, file = "../vignette_data/pbmc_multiomic.rds")
```


<details>
  <summary>**Session Info**</summary>

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

</details>