Reply
Highlighted

How to remove batch effects in biological replicates

Posted By: frank9, on Jan 31, 2018 at 7:49 PM

Hello 10Xers,

 

We are now measuring immune cell populations from 3 individuals. With cellranger pipeline, i.e., aggr, we merged the 3 samples into one and visualized in Loupe Cell Browser. We found some discrepancies between the 3 samples, so does celll ranger aggr consider the potential batch effect between samples? Or ERCC Spike in can provide a solution? We noticed some methods can remove the batch effect with the ERCC Spike such as SCDE.

 

Another question is since 10x is zero-inflated, should I preprocess the UMI matrix with software such as ZIFA before downstream analysis?

3 Replies

Re: How to remove batch effects in biological replicates

Posted By: Leah, on Feb 2, 2018 at 12:24 PM

Hello!

 

Currently, the cellranger aggr pipeline only performs read depth normalization among libraries/samples combined.

 

In general, normalization and correcting for batch effects in single cell RNA-seq data are areas of active research.  From the literature, there are a number of packages in R, including Seurat, scran, and scone, which attempt to address these issues (in addition to scde which you cited).

 

Below are examples of performing batch correction in Seurat:

 

  1. The following tutorial shows an example using the RegressOut function:
  2. Alternatively, below is a tutorial for “aligning” data sets/samples (which are expected to be similar):

With regards to the perception that 10x data is zero-inflated, there is an interesting blog post regarding this topic here:

http://www.nxn.se/valent/2017/11/16/droplet-scrna-seq-is-not-zero-inflated

 

At present, we don’t recommend any additional pre-processing of the data to adjust for zero counts.

Re: How to remove batch effects in biological replicates

Posted By: frank9, on Feb 5, 2018 at 12:02 AM

Thanks for your quick reply, the post is really interesting. I'll try the normalization method in Seurat and compare the results of both methods. By the way, Is there a suggestive technical way to remove multiplet in the results. Filtering cells with large UMI count may be a solution but are not practical in our research. Or biological experiment validation is the only solution?

 

Thanks,

Frank

Re: How to remove batch effects in biological replicates

Posted By: Leah, on Feb 6, 2018 at 10:46 AM

HI Frank,

 

Unfortunately, there is currently no automated way or algorithm in Cell Ranger to filter for multiplets in the Single Cell 3' assay data.  As you noted, filtering for large UMI counts may not always make sense given the range of RNA content variation in some samples.