Deconvolution_with_scaden_for_spatial_transcriptomics_data.Rmd

---
title: "Deconvolution with scaden for spatial transcriptomics data"
output: rmarkdown::html_vignette
date: 'Compiled: `r format(Sys.Date(), "%B %d, %Y")`'
vignette: >
  %\VignetteIndexEntry{Deconvolution with scaden for spatial transcriptomics data}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  tidy = TRUE,
  tidy.opts = list(width.cutoff = 95),
  message = FALSE,
  warning = FALSE, 
  error = FALSE,
  fig.align = "center"
)
```


Recent studies ([Zubair el.al.](https://academic.oup.com/nar/article/50/14/e80/6583238), [Wang el.al.](https://www.biorxiv.org/content/10.1101/2022.12.15.520612v1)) have indicated the potential to apply bulk-deconvolution methods to new technologies such as spatial transcriptomics when the resolution is not at single-cell level. 

For this tutorial, we will use `scaden` to deconvolute a dataset of sagital mouse brain slices generated using the Visium v1 chemistry, and generate cell-type specific gene expression. The data is provided by `Seurat` package and can be downloaded via `InstallData()`. If you haven't installed `scaden`, please follow instruction [here](https://liuy12.github.io/SCdeconR/).

# Load required libraries and data

```{r eval=FALSE}
## download data if it's not already installed
SeuratData::InstallData("stxBrain")
```

```{r}
library(SCdeconR)
library(plotly)
## load brain spatial data from sample anterior1
brain <- SeuratData::LoadData("stxBrain", type = "anterior1")
## proprocess data
brain <- SCTransform(brain, assay = "Spatial", verbose = FALSE)
## see ?refdata_brain for details
data("refdata_brain")
```

# Perform spot deconvolution using scaden
```{r echo=FALSE, message=FALSE, warning=FALSE}
phenodata <- refdata_brain@meta.data
## provide directory that contains python binary to pythonpath, e.g. /usr/bin/
decon_res <- scdecon(bulk = GetAssayData(brain, slot = "data", assay = "SCT"),
                     ref = GetAssayData(refdata_brain, slot = "data", assay = "SCT"),
                     phenodata = phenodata,
                     filter_ref = TRUE,
                     decon_method = "scaden",
                     norm_method = "none",
                     trans_method = "none",
                     marker_strategy = "all",
                     pythonpath ="/research/bsi/projects/PI/tertiary/Wang_Chen_m092469/s216340.Ovarian_cell_line_RNAseq/processing/personal_dir/python_venv/anaconda/envs/scaden_env/bin/python")
```


```{r eval=FALSE}
## provide directory that contains python binary to pythonpath, e.g. /usr/bin/
decon_res <- scdecon(bulk = GetAssayData(brain, slot = "data", assay = "SCT"),
                     ref = GetAssayData(refdata_brain, slot = "data", assay = "SCT"),
                     phenodata = refdata_brain@meta.data,
                     filter_ref = TRUE,
                     decon_method = "scaden",
                     norm_method = "none",
                     trans_method = "none",
                     marker_strategy = "all",
                     pythonpath ="/path/to/python/bin/dir/", 
                     tmpdir = "/path/to/tmp/dir/")
```

Visualize distribution of cell-type proprotions

```{r fig.width=8, fig.height=6}
prop_barplot(prop = decon_res[[1]], interactive = TRUE)
```

Add predicted cell-type proportions as new `assaydata` to `Seurat` object.

```{r}
prop_assay <- CreateAssayObject(data = decon_res[[1]])
brain[["PROP"]] <- prop_assay
```

Visualize predicted cell proportions for L5 IT, L6 IT and macrophage in spatial

```{r fig.width=12, fig.height=6}
DefaultAssay(brain) <- "PROP"
SpatialFeaturePlot(brain, features = c("L5.IT", "L6.IT"), pt.size.factor = 1.6, ncol = 2)
```

# Compute cell-type specific gene expression
```{r}
ct_exprs_list <- celltype_expression(bulk = GetAssayData(brain, slot = "data", assay = "SCT"),
                                     ref = GetAssayData(refdata_brain, slot = "data", assay = "SCT"),
                                     phenodata = refdata_brain@meta.data,
                                     prop = decon_res[[1]],
                                     UMI_min = 0,
                                     CELL_MIN_INSTANCE = 1)

## create assaydata and add to Seurat object
for(i in 1:length(ct_exprs_list)){
  ct_assay <- CreateAssayObject(data = as.matrix(ct_exprs_list[[i]]))
  ct_name <- names(ct_exprs_list)[i]
  brain[[ct_name]] <- ct_assay
}
```

We can then examine expression of genes using cell-type specific gene expression assays

```{r, fig.width=12, fig.height=6}
DefaultAssay(brain) <- "L5.IT"
SpatialFeaturePlot(brain, features = c("Rorb", "Fezf2"), pt.size.factor = 1.6, ncol = 2)
```