Answer the reader's question (7): About doublet

Several questions have been released on gongzong number , Next, I will give my thoughts in turn , For reference only . If you have different opinions , Welcome to email me to discuss ：[email protected]

problem

*

How do I get there? doublet？ When selecting a large category for dimension reduction clustering of small categories , Some people will still check various categories marker Expression of , Is this step necessary ？

*

1

The most effective way is to start from the upstream , Control the number of cells and optimize the process before sequencing .

2

However, those who analyze the data can only look at it from the perspective of analysis . Usually I will 「 First use software to predict 」, Please refer to a post I wrote before ： Single cell analysis record (4): doublet testing .

Here comes a question , Go to doublet At which step ：

• Run away cellranger Get the matrix , Go to doublet, basic QC( Gene number 、UMI Count and so on )...
• Run away cellranger Get the matrix , basic QC( Gene number 、UMI Count and so on ), Go to doublet...

There should be little difference between the two . I'm usually the first , Considering the basic QC The threshold may be changed back and forth several times , So basic QC This step should be relatively backward , Save running doublet This step

I usually use multiple software ,「 Two 」 The software also reports doublet Of cellular barcode, I will pick out and remove .

3

besides , I will also use some classic marker Come looking for doublet.（ This method can also be used to annotate cells ）

celltype_marker=c(
  "Epcam",# epithelial cells  epithelial
  "Pecam1","Cdh5",# Endothelial cells  endothelial
  "Pdgfra","Col1a1","Col3a1",# Fibroblasts  fibroblasts
  "Fcgr1","Cd163","Aif1","Cd68",# Myeloid cells  myeloid
  "Ms4a1",#B cells 
  "Cd3g","Cd3e",#T cells 
  "Ncr1",#NK cells 
  "Ptprc"# Immune cells 
)
VlnPlot(allseu,features = celltype_marker,pt.size = 0,ncol = 2)

Like this , If a group cluster It expresses what is unlikely to appear at the same time gene, It could be doublet. Of course, we should rely on some 「 Experience 」, For example, fibrous marker There may be 「 Certain extent 」 The expression of ,CD4( One gene ) There may be 「 Certain extent 」 The expression of , It's possible . but T Cell overexpression EPCAM, I don't believe it .

4

There is another experience ： use marker Identification of doublet When , Identified doublet( This is the premise. ), They may gather together to form a small group , It may also look like a bridge connecting two subgroups .

5

in addition , All of the above are for general categories clustering Description of , Do small classes reclustering When , I generally don't think about going between classes doublet, First, it is difficult to distinguish , The second is between subclasses doublet Formation probability ratio class doublet The probability of formation is much lower .

6

There's another situation ( Relatively rare ), Between samples doublet. At this time , utilize SNP It can be distinguished accurately sample1, sample2, as well as doublet(sample1+sample2).

There are already some tools available from scRNA-seq Get a small amount of data SNP Information , And achieve the above purpose , such as Souporcell( original text https://www.nature.com/articles/s41592-020-0820-1), The accuracy is OK .

But in most cases, one sample is tested , This idea is not feasible .

Back to the first 2 A little question

*

When selecting a large category for dimension reduction clustering of small categories , Some people will still check various categories marker Expression of , Is this step necessary ？

*

It is necessary to . Doing it reclustering At this stage , Inspection category marker The expression of , I often see other kinds of cells ( Or it is composed of this kind of cells and other large types of cells doublet) Enter in disorder . Friends who have not done this step before , You can try , There will be surprises ( scared ).

At this time, the most rigorous way is to put other large types of cells back where they should go , Single cell clustering annotation is a “ Opening and closing ” The process of , It is also the most time-consuming and cumbersome process .

If the number is small , Directly remove , It doesn't matter much .

If it is composed of this kind of cells and other large types of cells doublet, Directly remove .