06_Combined_Integration.Rmd

---
title: "Combined Integration"
output:
    workflowr::wflow_html:
        code_folding: hide
---

```{r knitr, include = FALSE}
DOCNAME = "06_Combined_Integration"
knitr::opts_chunk$set(autodep        = TRUE,
                      cache          = TRUE,
                      cache.path     = paste0("cache/", DOCNAME, "/"),
                      cache.comments = FALSE,
                      echo           = TRUE,
                      error          = FALSE,
                      fig.align      = "center",
                      fig.width      = 10,
                      fig.height     = 8,
                      message        = FALSE,
                      warning        = FALSE)
```

```{r libaries, cache = FALSE}
# scRNA-seq
library("SingleCellExperiment")
library("Seurat")

# Paths
library("here")

# Tidyverse
library("tidyverse")
```

```{r source, cache = FALSE}
source(here("R/seurat_helpers.R"))
```

```{r paths}
org123_path <- here("data/processed/Organoid123_SCE_filtered.Rds")
org4_path <- here("data/processed/Organoid4_SCE_filtered.Rds")
lind_path <- here("data/processed/Lindstrom_SCE_filtered.Rds")
```

Introduction
============

Here we will combine the two batches of organoid samples with the Lindstrom
data using Seurat's integration method.

```{r load-org123, cache.extra = tools::md5sum(org123_path)}
if (file.exists(org123_path)) {
    org123_sce <- read_rds(org123_path)
} else {
    stop("Organoid123 dataset is missing. ",
         "Please run '01_Organoid123_QC.Rmd' first.",
         call. = FALSE)
}
```

```{r load-org4, cache.extra = tools::md5sum(org4_path)}
if (file.exists(org4_path)) {
    org4_sce <- read_rds(org4_path)
} else {
    stop("Organoid4 dataset is missing. ",
         "Please run '02_Organoid4_QC.Rmd' first.",
         call. = FALSE)
}
```

```{r load-lind, cache.extra = tools::md5sum(lind_path)}
if (file.exists(lind_path)) {
    lind_sce <- read_rds(lind_path)
} else {
    stop("Lindstrom dataset is missing. ",
         "Please run '05_Lindstrom_QC.Rmd' first.",
         call. = FALSE)
}
```

The datasets are currently in `SingleCellExperiment` format but to use Seurat we
need to convert them to `Seurat` format.

```{r convert}
org123 <- as.seurat(org123_sce)
org4 <- as.seurat(org4_sce)
lind <- as.seurat(lind_sce)
```

Variable genes
==============

Before we can align the datasets we need to select a set of genes to use. We
do this by looking for highly variable genes.

```{r var-genes}
org123 <- NormalizeData(org123, display.progress = FALSE)
org4 <- NormalizeData(org4, display.progress = FALSE)
lind <- NormalizeData(lind, display.progress = FALSE)

org123 <- ScaleData(org123, display.progress = FALSE)
org4 <- ScaleData(org4, display.progress = FALSE)
lind <- ScaleData(lind, display.progress = FALSE)

org123 <- FindVariableGenes(org123, do.plot = FALSE)
org4 <- FindVariableGenes(org4, do.plot = FALSE)
lind <- FindVariableGenes(lind, do.plot = FALSE)

genes.use <- unique(c(org123@var.genes, org4@var.genes, lind@var.genes))
genes.use <- intersect(genes.use, rownames(org123@scale.data))
genes.use <- intersect(genes.use, rownames(org4@scale.data))
genes.use <- intersect(genes.use, rownames(lind@scale.data))
```

Seurat finds `r length(org123@var.genes)` highly variable genes for the first
organoid dataset and `r length(org4@var.genes)` highly variable genes for
the second organoid dataset. There are `r length(lind@var.genes)` highly
variable genes for the Lindstrom dataset. We take these genes and use those that
are present in both datasets. This leaves us with `r length(genes.use)` selected 
genes.

Merging
=======

Before we try to properly align the datasets let's see what happens if we just 
naively merge them together. If this works we might not need to do anything
more sophisticated.

```{r merge, cache.lazy = FALSE}
merged <- MergeSeurat(org123, org4, 
                      add.cell.id1 = "Org123", add.cell.id2 = "Org4")
merged <- MergeSeurat(merged, lind, add.cell.id2 = "Lind")
merged <- ScaleData(merged, display.progress = FALSE)
merged <- RunPCA(merged, pc.genes = genes.use, pcs.compute = 20,
                 do.print = FALSE)
merged <- ProjectPCA(merged, do.print = FALSE)
merged <- RunTSNE(merged, dims.use = 1:20, do.fast = TRUE)
merged@meta.data$DatasetSample <- paste(merged@meta.data$Dataset,
                                        merged@meta.data$Sample, sep = "_")
TSNEPlot(merged, group.by = "DatasetSample")
```

Ideally we would see an even mix of the two datasets here but there is some
pretty clear separation. Let's move on to the alignment process.

Perform CCA
===========

The first step of Seurat's alignment is to perfrom Canoical Correlation Analysis
(CCA). This combines the datasets in the same multi-dimensional space.

```{r cca, results = "hide", cache.lazy = FALSE}
combined <- RunMultiCCA(list(prefixCellNames(org123, "Org123"),
                             prefixCellNames(org4, "Org4"),
                             prefixCellNames(lind, "Lind")),
                        genes.use = genes.use, num.cc = 50)

p1 <- DimPlot(combined, reduction.use = "cca", group.by = "Dataset", 
              pt.size = 0.5, do.return = TRUE)
p2 <- VlnPlot(combined, features.plot = "CC1", group.by = "Dataset", 
              do.return = TRUE)
plot_grid(p1, p2)
```

Choose CCs
==========

Now we need to choose the dimensions to use for aligning the datasets. This is
similar to choosing principal components for clustering a single dataset. Here
for look for a drop off in correlation strength for each CC.

```{r plot-CCs, cache.lazy = FALSE}
plot <- MetageneBicorPlot(combined, grouping.var = "Dataset", dims.eval = 1:50, 
                          display.progress = FALSE)
```

We can also look at the genes associated with each of the first nine CCs.

```{r CC-heatmaps, fig.height = 10}
DimHeatmap(object = combined, reduction.type = "cca", cells.use = 500, 
           dim.use = 1:9, do.balanced = TRUE)
```

```{r n-dims}
n.dims <- 20
```

Based on these plots we will to use the first `r n.dims` CCs.

Align CCA subspaces
===================

Now we can align the CCA subspaces for the two datasets. This makes the
dimensions more comparable in a way we can use for clustering.

```{r align, cache.lazy = FALSE}
combined <- AlignSubspace(combined, reduction.type = "cca",
                          grouping.var = "Dataset", dims.align = 1:n.dims,
                          verbose = FALSE)

p1 <- VlnPlot(object = combined, features.plot = "ACC1", group.by = "Dataset", 
              do.return = TRUE)
p2 <- VlnPlot(object = combined, features.plot = "ACC2", group.by = "Dataset", 
              do.return = TRUE)
plot_grid(p1, p2)
```

```{r align-tSNE, cache.lazy = FALSE}
combined <- RunTSNE(combined, reduction.use = "cca.aligned", dims.use = 1:20)
combined@meta.data$DatasetSample<- paste(combined@meta.data$Dataset,
                                         combined@meta.data$Sample,
                                         sep = "_")
TSNEPlot(combined, pt.size = 0.5, group.by = "DatasetSample")
```

This looks to have worked well, there is no obvious separation between the two
datasets.

This combined dataset of `r nrow(combined@scale.data)` genes and
`r ncol(combined@scale.data)` cells is the starting point for our clustering
analysis.

```{r write-combined}
write_rds(combined, here("data/processed/Combined_Seurat.Rds"))
```

Metadata
========