I am interested in using Seurat to compare wild type vs Mutant. I don't know how to use the package. How can I test whether mutant mice, that have deleted gene, cluster together?
$\begingroup$ This question is too vague and open-ended for anyone to give you specific help, right now. Can you provide a bit more detail? Have you run Seurat on your WT samples and have well-defined clusters that you want to compare against in your mutant samples? If you can show some figures or briefly describe what you've done so far, or specifically what you're trying to achieve, it will be much easier for someone to provide guidance. $\endgroup$– James HawleyNov 14, 2018 at 15:07
Single-cell analysis to compare samples is a long a difficult process. There is very good documentation for 10x Genomics cellranger, the DropSeq Pipeline and the Seurat R package. These tools all have GitHub repositories and the authors are very responsive if you encounter issues.
Depending on the technology used to generate the data, you'll need to use either cellranger or DropSeq to process the FASTQ files. These are designed to account for the the different experimental design. Still, it is important to ensure that your samples have been processed in the same way to reduce batch effects.
This question is to open-ended to cover all of the details needed to perform the analysis but here is some of the main things to consider and the general process to perform such as analysis. You will need competence using the command-line (shell) and programming in R (or Python) to perform single-cell analysis. I recommend performing this analysis on a remote server as some steps are memory intensive.
The general overview to compare samples is:
Demultiplex the reads for different indicies. You need to do this even if you've only put one sample per lane (to remove the index from the reads).
cellranger mkfastqor Illumina's
bcl2fastqwill do this.
Obtain a reference genome (FASTA) and gene annotation (GTF) for the species you are working with. You can prepare a reference transcriptome with
cellranger countor the DropSeq pipeline on each sample separately. These will both perform STAR (splice-aware) alignment of paired-end RNA-Seq reads and count UMIs for each cell-barcode and gene. Cells will be identified and filtered by an automatic UMI threshold (this can be changed by forcing with expected number of cells). This will return a gene-barcode matrix.
You need to aggregate results of cellranger runs for different samples with
cellranger aggr. This will perform downsampling by default to normalise the number of reads between samples. It will also perform graph-based (Louvian) clustering on the combined samples and return HTML and Cloupe summary data to explore, in addition to a combined gene-barcode matrix for all samples.
Single-cell RNA-Seq experiments are subject to batch effects. The CCA method from
Seuratand the MNN method from
scranare both available in the respective R packages to account for this (with different approaches). Batch effect correction needs to be performed before downstream analysis to ensure comparisons are valid.
This methods require an overlap between the samples to estimate technical errors and account for them. You need to bear in mind that neither of these methods works unless:
some of the cells detected in each sample are expected to be the same cell type
biological replicates of each sample have been performed
Once you have performed this correction, then standard
Seurat workflows such as those shown in the tutorials can be performed which will identify clusters. You can use these to identify clusters specific to each genotype. Bear in mind that some overlap should be expected and if there isn't one, batch effect correction will over-correct for this.
$\begingroup$ Hi Tom, Thank you for your advice. You have mentioned that we can use a remote server. Yes I have tried that. But I don't know how to save our data in a remote server. Can you tell me how to? Thank you so much! $\endgroup$– huaNov 18, 2018 at 20:59
$\begingroup$ CCA corrects based on shared variability. ie. if you expect to find a unique new cell type or cell state for your mutant condition than after CCA that cell type will more often than not be blended in with the other clusters. You won't detect it. If this is what you expect MNN might be better. However if you expect changes in gene expression between the same cluster for wt and mutant CCA might work just fine. Those limitations have been addressed very recently by many groups. $\endgroup$ Nov 18, 2018 at 22:31
$\begingroup$ MNN has similar limitations. It’s stated as such in the original publication of this method. $\endgroup$ Nov 18, 2018 at 23:55
$\begingroup$ You just need to run cellranger, it will write output files to disk. You save data to disk from R on a remote server. This is really a separate question and I think you’ll find answers for how to do it elsewhere. You need to ask specific questions in SE to get the best answers. $\endgroup$ Nov 18, 2018 at 23:59
There is no single ended answer to your question as everything will depend on your data. For example: Is this unsupervised data where you have to browse to 100 cell types for each sample or did you FACS sort on specific markers and are you looking for specific differences between 2 specific cell types etc...
Below you can find 4 recent packages that assist in your question. I strongly suggest you read the preprint of each, play around with your data and then follow the suggestion in the comments and ask more specific questions about how to proceed correctly.
If you want to use Seurat you can start here: Seurat 3.0 pre-release to integrate your data sets: https://satijalab.org/seurat/pancreas_integration_label_transfer.html
Seurat 2 tutorial on CCA to compare different conditions you can use these methods on your integrated data sets. https://satijalab.org/seurat/immune_alignment.html
LIGER Package and preprint can be found here https://macoskolab.github.io/liger/