---
title: "PROCESSING (II) Student's version"
date: "2023-11-05.10"
author:
  - name: "YOU"
    email: "you@you.you"
output:
  html_document: 
    highlight: true  ## Theme for the code chunks
    number_sections: true  ## Adds number to headers (sections)
    theme: flatly  ## CSS theme for the HTML page
    toc: false  ## Adds a table of content
    self_contained: true ## Includes all plots/images within the HTML
    code_folding: show
    thumbnails: false
    lightbox: true
    fig_caption: false
    gallery: true
    use_bookdown: true
always_allow_html: true ## Allow plain HTML code in the Rmd
---

This file describes the different steps to perform first part of data processing for the single cell RNAseq data analysis training course for the EBAII n1 2023, covering these steps :

* Reduction dimension of the expression data
* Visualization targeted at a human brain

# Load the latest Seurat object

Start a **Rstudio** session

First, we will set some parameters

```{r seed}
## Fixed seed
my_seed <- 1337

## Your project name on the IFB cluster (CHANGE IT FOR YOURS !)
my_project <- 'golf'
```

You can load the latest Seurat object you saved as a RDS at the preceding step (Proc.1 : normalization with LN, and scaling using 2000 HVGs) :

```{r read_trick}
sobj <- readRDS(file = paste0('/shared/projects/', my_project, '/TD/RESULTS/TD3A_02_Scaled.2K.RDS'))
```

<font size="4">***OR***</font>_, if you did not (could not) create it earlierly, you can load it from here :_

```{r read_safe}
sobj <- readRDS(file = '/shared/projects/2325_ebaii/SingleCell/TD_DATA/DATA_START/DIMENSION_REDUCTION/TD3A_02_Scaled.2K.RDS')
```

# Dimension reduction

This step originates from the observation that we do not want nor need to characterize **each** of our **thousends of cells**, but **groups** of them (clusters ? cell types ? other ?). Thus, we do no need all data, and even may benefit from such a reduction :

* Reduce the data complexity
  * For interpretation
  * For computations
* Increase the quality of information contained in the data
  * **Enriching** "good biological **signals**"
  * **Discarding noise** / cell-specific signals

There is a **multitude of methods** for dimension reduction

Here, we will use the grand-mother of all : the PCA (Principal Component Analysis)

_But how ? (did you see it coming ?)_

```{r h_RunPCA, class.source = "fold-hide"}
?Seurat::RunPCA
```

***Question 1*** : How many principal components (PC) will be generated by default ?

***Question 2*** : Which data type (ie, which Seurat object ***slot***) will be used to generate the components ?

Perform PCA on our data

```{r PCA}
sobj <- Seurat::RunPCA(
  object = sobj, 
  assay = 'RNA', 
  seed.use = my_seed, 
  verbose = FALSE)
```

Visualization of the very first two components :

```{r PCAplot, fig.align='center'}
Seurat::DimPlot(
  object = sobj, 
  reduction = 'pca',
  dims = c(1,2), 
  group.by = 'CC_Seurat_Phase')
```

***Question 1*** : Give me your interpretation / feelings from this plot !

***Question 2*** : Should we stop at using 2 dimensions to interpret our data ?

Description :

```{r PCAdesc}
EBAII.n1.SC.helper::seurat4_descriptor(sobj = sobj, describe = 'dimred')
```

* Maybe we shoud reduce information a tad more, just for the sake of ... 
* ... understanding our data ...
* ...with our poooooor human brains ...
* ... born and raised in a 3D euclidean world.

# Visualization

This final processing step required to finaly **observe** our data requires a novel dimension reduction method with a very high challeng to overcome : reduce a space of dozens of dimensions to just 2, or 3 !

We will use the UMAP method.

_**BUT HOW ?** (I'm pretty sure you saw it coming this time)_

```{r, class.source = "fold-hide"}
?Seurat::RunUMAP
```

## Selecting dimensions

So, we now have to choose the number of PCA dimensions to use for this UMAP reduction.

***Question*** : Do you have an idea of this number ?

We will use a very simple graphical method : the observation of the amount of global variance explained by each component.

```{r h_elbow}
?Seurat::ElbowPlot
```

```{r elbow, fig.align='center'}
Seurat::ElbowPlot(
  object = sobj, 
  ndims = 50)
```

***Question*** : Any more precise idea ?

## Assessing dimensions

To demonstrate the effect of the number of PC dimensions used as input to the UMAP generation, I will perform a DEMO using 4 different PC values : **3, 7, 23 and 49**.


**3 PCs**

```{r umap3}
sobj <- Seurat::RunUMAP(
  object = sobj, assay = 'RNA', 
  graph.name = 'RNA_snn', 
  reduction = 'pca', dims = 1:3, 
  seed.use = my_seed)


Seurat::DimPlot(
  object = sobj, 
  reduction = 'umap')
```

**7 PCs**

```{r umap7}
sobj <- Seurat::RunUMAP(
  object = sobj, assay = 'RNA', 
  graph.name = 'RNA_snn', 
  reduction = 'pca', dims = 1:7, 
  seed.use = my_seed)


Seurat::DimPlot(
  object = sobj, 
  reduction = 'umap')
```

**25 PCs**

```{r umap25}
sobj <- Seurat::RunUMAP(
  object = sobj, assay = 'RNA', 
  graph.name = 'RNA_snn', 
  reduction = 'pca', dims = 1:25, 
  seed.use = my_seed)


Seurat::DimPlot(
  object = sobj, 
  reduction = 'umap')
```

**49 PCs**

```{r umap49}



## OH NO !!
## SOMEONE DELETED THE Seurat::RunUMAP COMMAND OF THAT CHUNK :(
## IT'S UP TO YOU :)



Seurat::DimPlot(
  object = sobj, 
  reduction = 'umap')
```

***Question*** : Your conclusion ?

You can now perform a final UMAP with the PC dimensions of your choice.

For the next steps of the training, I used 20 dimensions, but we will stop using TD3A solely for now.

_**Towards level 2** : Try creating your chunk for 20 dims._

# Save

You can now save the results of your hard work :

```{r save}
saveRDS(object = sobj, 
        file = paste0('/shared/projects/', my_project, '/TD/RESULTS/TD3A_03_DimRed.RDS'), 
        compress = 'bzip2')
```


_Rsession_

```{r rsession}
utils::sessionInfo()
```