How do I pick thresholds and parameters in my analysis? Whatâs a âreasonableâ number, and will the world collapse if I pick the wrong one?
Objectives:
Interpret quality control plots to direct parameter decisions
Repeat analysis from matrix to clustering
Identify decision-making points
Appraise data outputs and decisions
Explain why single cell analysis is an iterative (i.e. the first plots you generate are not final, but rather you go back and re-analyse your data repeatedly) process
Youâve done all the work to make a single cell matrix, with gene counts and mitochondrial counts and buckets of cell metadata from all your variables of interest. Now itâs time to fully process our data, to remove low quality cells, to reduce the many dimensions of data that make it difficult to work with, and ultimately to try to define our clusters and to find our biological meaning and insights! There are many packages for analysing single cell data - Seurat Satija et al. 2015, Scanpy Wolf et al. 2018, Monocle Trapnell et al. 2014, Scater McCarthy et al. 2017, and so forth. Weâre working with Scanpy, because currently Galaxy hosts the most Scanpy tools of all of those options.
Comment: Tutorials everywhere?
This tutorial is similar to another fantastic tutorial: Clustering 3k PBMC. That tutorial will go into much further depth on the analysis, in particular the visualisation and science behind identifying marker genes. Their experimental data is clean and well annotated, which illustrates the steps beautifully. Here, we work more as a case study with messier data, to help empower you in making choices during the analysis. We highly recommend you work through all the galaxy single cell tutorials to build confidence and expertise! For trainers, note that there are small-group options in this tutorial.
Weâve provided you with experimental data to analyse from a mouse dataset of fetal growth restriction Bacon et al. 2018. This is the full dataset generated from this tutorial if you used the full FASTQ files rather than the subsampled ones (see the study in Single Cell Expression Atlas and the project submission). You can find this dataset in this input history or download from Zenodo below.
You can access the data for this tutorial in multiple ways:
Your own history - If youâre feeling confident that you successfully ran a workflow on all 7 samples from the previous tutorial, and that your resulting 7 AnnData objects look right (you can compare with the answer key history), then you can use those! Be careful, in the previous tutorials we worked on downsampled datasets, but here we start with the object containing total data from all the samples! To avoid a million-line history, I recommend dragging the resultant datasets into a fresh history:
There 3 ways to copy datasets between histories
From the original history
Click on the galaxy-gear icon which is on the top of the list of datasets in the history panel
Click on Copy Datasets
Select the desired files
Give a relevant name to the âNew historyâ
Validate by âCopy History Itemsâ
Click on the new history name in the green box that have just appear to switch to this history
Using the galaxy-columnsShow Histories Side-by-Side
Click on the galaxy-dropdown dropdown arrow top right of the history panel (History options)
Click on galaxy-columnsShow Histories Side-by-Side
If your target history is not present
Click on âSelect historiesâ
Click on your target history
Validate by âChange Selectedâ
Drag the dataset to copy from its original history
Drop it in the target history
From the target history
Click on User in the top bar
Click on Datasets
Search for the dataset to copy
Click on its name
Click on Copy to current History
Importing from a history - You can import this history
Open the link to the shared history
Click on the Import this history button on the top left
Click galaxy-uploadUpload Data at the top of the tool panel
Select galaxy-wf-editPaste/Fetch Data
Paste the link(s) into the text field
Press Start
Close the window
Renamegalaxy-pencil the datasets Mito-counted AnnData
Check that the datatype is h5ad
Click on the galaxy-pencilpencil icon for the dataset to edit its attributes
In the central panel, click galaxy-chart-select-dataDatatypes tab on the top
In the galaxy-chart-select-dataAssign Datatype, select h5ad from âNew typeâ dropdown
Tip: you can start typing the datatype into the field to filter the dropdown menu
Click the Save button
Importing data from EBI Single Cell Expression Atlas
You can also pull the data from publicly available Single Cell Expression Atlas. You can simply access the dataset we are working on by using the tool EBI SCXA Data Retrieval with experiment id of E-MTAB-6945. This short tutorial will show you how to use this tool and modify the output so that itâs compatible with this tutorial and its workflow.
Important tips for easier analysis
Tools are frequently updated to new versions. Your Galaxy may have multiple versions of the same tool available. By default, you will be shown the latest version of the tool. This may NOT be the same tool used in the tutorial you are accessing. Furthermore, if you use a newer tool in one step, and try using an older tool in the next step⊠this may fail! To ensure you use the same tool versions of a given tutorial, use the Tutorial mode feature.
Open your Galaxy server
Click on the curriculum icon on the top menu, this will open the GTN inside Galaxy.
Navigate to your tutorial
Tool names in tutorials will be blue buttons that open the correct tool for you
Note: this does not work for all tutorials (yet)
You can click anywhere in the grey-ed out area outside of the tutorial box to return back to the Galaxy analytical interface
Warning: Not all browsers work!
Weâve had some issues with Tutorial mode on Safari for Mac users.
Try a different browser if you arenât seeing the button.
Did you know we have a unique Single Cell Omics Lab with all our single cell tools highlighted to make it easier to use on Galaxy? We recommend this site for all your single cell analysis needs, particularly for newer users.
The Single Cell Omics Lab is a different view of the underlying Galaxy server that organises tools and resources better for single-cell users! It also provides a platform for communities to engage and connect; distribute more targeted news and events; and highlight community-specific funding sources.
When something goes wrong in Galaxy, there are a number of things you can do to find out what it was. Error messages can help you figure out whether it was a problem with one of the settings of the tool, or with the input data, or maybe there is a bug in the tool itself and the problem should be reported. Below are the steps you can follow to troubleshoot your Galaxy errors.
Expand the red history dataset by clicking on it.
Sometimes you can already see an error message here
View the error message by clicking on the bug icongalaxy-bug
Check the logs. Output (stdout) and error logs (stderr) of the tool are available:
Expand the history item
Click on the details icon
Scroll down to the Job Information section to view the 2 logs:
You have generated an annotated AnnData object from your raw scRNA-seq fastq files. However, you have only completed a âroughâ filter of your dataset - there will still be a number of âcellsâ that are actually just background from empty droplets or simply low-quality. There will also be genes that could be sequencing artifacts or that appear with such low frequency that statistical tools will fail to analyse them. This background garbage of both cells and genes not only makes it harder to distinguish real biological information from the noise, but also makes it computationally heavy to analyse. These spurious reads take a lot of computational power to analyse! First on our agenda is to filter this matrix to give us cleaner data to extract meaningful insight from, and to allow faster analysis.
Question
What information is stored in your AnnData object? The last tool to generate this object counted the mitochondrial associated genes in your matrix. Where is that data stored?
While you are figuring that out, how many genes and cells are in your object?
You want to use the same tool you used in the previous tutorial to examine your AnnData, since itâs not necessarily as simple as examining the Anndata dataset in the history!
Hands On: Inspecting AnnData Objects
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âWhat to inspect?â: General information about the object
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âWhat to inspect?â: Key-indexed observations annotation (obs)
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âWhat to inspect?â: Key-indexed annotation of variables/features (var)
If you examine your AnnData object, youâll find a number of different quality control metrics for both cells toolobs and tool genes var.
For instance, you can see a n_cells under var, which counts the number of cells that gene appears in.
In the obs, you have both discrete and log-based metrics for n_genes, how many genes are counted in a cell, and n_counts, how many UMIs are counted per cell. So, for instance, you might count multiple GAPDHs in a cell. Your n_counts should thus be higher than n_genes.
But what about the mitochondria?? Within the cells information obs, the total_counts_mito, log1p_total_counts_mito, and pct_counts_mito has been calculated for each cell.
You can see in the toolGeneral information about the object output that the matrix is 31178 x 35734. This is obs x vars, or rather, cells x genes, so there are 31178 cells and 35734 genes in the matrix.
timeTop time-saving advice - turn the 3 Inspect AnnData outputs above into a workflow for quick access!
Clean up your history: remove any failed (red) jobs from your history by clicking on the galaxy-delete button.
This will make the creation of the workflow easier.
Click on galaxy-gear (History options) at the top of your history panel and select Extract workflow.
The central panel will show the content of the history in reverse order (oldest on top), and you will be able to choose which steps to include in the workflow.
Replace the Workflow name to something more descriptive.
Rename each workflow input in the boxes at the top of the second column.
If there are any steps that shouldnât be included in the workflow, you can uncheck them in the first column of boxes.
Click on the Create Workflow button near the top.
You will get a message that the workflow was created.
Generate QC Plots
We want to filter our cells, but first we need to know what our data looks like. There are a number of subjective choices to make within scRNA-seq analysis, for instance we now need to make our best informed decisions about where to set our thresholds (more on that soon!). Weâre going to plot our data a few different ways. Different bioinformaticians might prefer to see the data in different ways, and here we are only generating some of the myriad of plots you can use. Ultimately you need to go with what makes the most sense to you.
timeTop time-saving advice - turn the following QC plots into a workflow so you can re-run it easily throughout analysing your own data!.
Creating the plots
Hands On: Making QC plots
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: genotype
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: sex
Renamegalaxy-pencil output Violin - sex - log
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: batch
Renamegalaxy-pencil output Violin - batch - log
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Scatter plot along observations or variables axes, using 'pl.scatter'
âPlotting tool that computed coordinatesâ: Using coordinates
âx coordinateâ: log1p_total_counts
ây coordinateâ: pct_counts_mito
Renamegalaxy-pencil output Scatter - mito x UMIs
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Scatter plot along observations or variables axes, using 'pl.scatter'
âPlotting tool that computed coordinatesâ: Using coordinates
âx coordinateâ: log1p_n_genes_by_counts
ây coordinateâ: pct_counts_mito
Renamegalaxy-pencil output Scatter - mito x genes
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-counted AnnData
âMethod used for plottingâ: Generic: Scatter plot along observations or variables axes, using 'pl.scatter'
âPlotting tool that computed coordinatesâ: Using coordinates
âx coordinateâ: log1p_total_counts
ây coordinateâ: log1p_n_genes_by_counts
âColor byâ: pct_counts_mito
Renamegalaxy-pencil output Scatter - genes x UMIs
Analysing the plots
Thatâs a lot of information! Letâs attack this in sections and see what questions these plots can help us answer. The scratchbookgalaxy-scratchbook may help here to look at the different plots at the same time!
If you would like to view two or more datasets at once, you can use the Window Manager feature in Galaxy:
Click on the Window Manager icon galaxy-scratchbook on the top menu bar.
You should see a little checkmark on the icon now
Viewgalaxy-eye a dataset by clicking on the eye icon galaxy-eye to view the output
You should see the output in a window overlayed over Galaxy
You can resize this window by dragging the bottom-right corner
Click outside the file to exit the Window Manager
Viewgalaxy-eye a second dataset from your history
You should now see a second window with the new dataset
This makes it easier to compare the two outputs
Repeat this for as many files as you would like to compare
You can turn off the Window Managergalaxy-scratchbook by clicking on the icon again
Question: Batch Variation
Are there differences in sequencing depth across the samples?
Which plot(s) addresses this?
How do you interpret it?
The plot violin - batch - log will have what youâre looking for!
Open image in new tab
Figure 1: Violin - batch - log (Raw)
Keeping in mind that this is a log scale - which means that small differences can mean large differences - the violin plots probably look pretty similar.
N703 and N707 might be a bit lower on genes and counts (or UMIs), but the differences arenât catastrophic.
The pct_counts_mito looks pretty similar across the batches, so this also looks good.
Nothing here would cause us to eliminate a sample from our analysis, but if you see a sample looking completely different from the rest, you would need to question why that is and consider eliminating it from your experiment!
Question: Biological Variables
Are there differences in sequencing depth across sex? Genotype?
Which plot(s) addresses this?
How do you interpret the sex differences?
How do you interpret the genotype differences?
Similar to above, the plots violin - sex - log and violin - genotype - log will have what youâre looking for!
Open image in new tab
There isnât a major difference in sequencing depth across sex, I would say - though you are welcome to disagree!
It is clear there are far fewer female cells, which makes sense given that only one sample was female. Note - that was an unfortunate discovery made long after generating libraries. Itâs quite hard to identify the sex of a neonate in the lab! In practice, try hard to not let such a confounding factor into your data! You could consider re-running all the following analysis without that female sample, if you wish.
In Violin - genotype - log, however, we can see there is a difference. The knockout samples clearly have fewer genes and counts. From an experimental point of view, we can consider, does this make sense?
Would we biologically expect that those cells would be smaller or having fewer transcripts? Possibly, in this case, given that these cells were generated by growth restricted neonatal mice, and in which case we donât need to worry about our good data, but rather keep this in mind when generating clusters, as we donât want depth to define clusters, we want biology to!
On the other hand, it may be that those cells didnât survive dissociation as well as the healthy ones (in which case weâd expect higher mitochondrial-associated genes, which we donât see, so we can rule that out!).
Maybe we unluckily poorly prepared libraries for specifically those knockout samples. There are only three, so maybe those samples are under-sequenced.
So what do we do about all of this?
Ideally, we consider re-sequencing all the samples but with a higher concentration of the knockout samples in the library. Any bioinformatician will tell you that the best way to get clean data is in the lab, not the computer! Sadly, absolute best practice isnât necessarily always a realistic option in the lab - for instance, that mouse line was long gone! - so sometimes, we have to make the best of it. There are options to try and address such discrepancy in sequencing depth. Thus, weâre going to take these samples forward and see if we can find biological insight despite the technical differences.
Now that weâve assessed the differences in our samples, we will look at the libraries overall to identify appropriate thresholds for our analysis.
Question: Filter Thresholds: genes
What threshold should you set for log1p_n_genes_by_counts?
Which plot(s) addresses this?
What number would you pick?
Any plot with log1p_n_genes_by_counts would do here, actually! Some people prefer scatterplots to violins.
In Scatter - mito x genes you can see how cells with log1p_n_genes_by_counts up to around, perhaps, 5.7 (around 300 genes) often have high pct_counts_mito.
You can plot this as just n_counts and see this same trend at around 300 genes, but with this data the log format is clearer so thatâs how weâre presenting it.
You could also use the violin plots to come up with the threshold, and thus also take batch into account. Itâs good to look at the violins as well, because you donât want to accidentally cut out an entire sample (i.e. N703 and N707).
Some bioinformaticians would recommend filtering each sample individually, but this is difficult in larger scale and in this case (youâre welcome to give it a go! Youâd have to filter separately and then concatenate), it wonât make a notable difference in the final interpretation.
Question: Filter Thresholds: UMIs
What threshold should you set for log1p_total_counts?
Which plot(s) addresses this?
What number would you pick?
As before, any plot with log1p_total_counts will do! Again, weâll use a scatterplot here, but you can use a violin plot if you wish!
We can see that we will need to set a higher threshold (which makes sense, as youâd expect more UMIâs per cell rather than unique genes!). Again, perhaps being a bit aggressive in our threshold, we might choose 6.3, for instance (which amounts to around 500 counts/cell).
In an ideal world, youâll see a clear population of real cells separated from a clear population of debris. Many samples, like this one, are under-sequenced, and such separation would likely be seen after deeper sequencing!
Question: Filter Thresholds: mito
What threshold should you set for pct_counts_mito?
Which plot(s) addresses this?
What number would you pick?
Any plot with pct_counts_mito would do here, however the scatterplots are likely the easiest to interpret. Weâll use the same as last time.
We can see a clear trend wherein cells that have around 5% mito counts or higher also have far fewer total counts. These cells are low quality, will muddy our data, and are likely stressed or ruptured prior to encapsulation in a droplet. While 5% is quite a common cut-off, this is quite messy data, so just for kicks weâll go more aggressive with a 4.5%.
In general, you must adapt all cut-offs to your data - metabolically active cells might have higher mitochondrial RNA in general, and you donât want to lose a cell population because of a cut-off.
Apply the thresholds
Itâs now time to apply these thresholds to our data! First, a reminder of how many cells and genes are in your object: 31178 cells and 35734 genes. Letâs see how that changes each time!
If you are working in a group, you can now divide up a decision here with one control and the rest varied numbers so that you can compare results throughout the tutorials.
Control
log1p_n_genes_by_counts > 5.7
log1p_total_counts > 6.3
pct_counts_mito < 4.5%
Everyone else: Choose your own thresholds and compare results!
Hands On: Filter cells by log1p_n_genes_by_counts
Scanpy FilterCells ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Mito-counted AnnData
In âParameters to select cells to keepâ:
param-repeatâInsert Parameters to select cells to keepâ
âName of parameter to filter onâ: log1p_n_genes_by_counts
âMin valueâ: 5.7
âMax valueâ: 20.0
Renamegalaxy-pencil output as Genes-filtered Object
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Genes-filtered Object
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: genotype
Renamegalaxy-pencil output Violin - Filterbygenes
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Genes-filtered Object
âWhat to inspect?â: General information about the object
Renamegalaxy-pencil output General - Filterbygenes
Note that the toolScanpy Filtercells allows you to put param-repeat multiple parameters at the same time (i.e. filter log1p_total_counts, log1p_n_genes_by_counts,and pct_counts_mito) in the same step. The only reason we arenât doing that here is so you can see what each filter accomplishes. As such, examine your plot and general information.
Question
Interpret the violin plot
How many genes & cells do you have in your object now?
The only part that seems to change is the log1p_n_genes_by_counts. You can see a flatter bottom to the violin plot - this is the lower threshold set. Ideally, this would create a beautiful violin plot because there would be a clear population of low-gene number cells. Sadly not the case here, but still a reasonable filter.
In General - Filterbygenes, you can see you now have 17,040 cells x 35,734 genes.
Hands On: Filter cells by log1p_total_counts
Scanpy FilterCells ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Genes-filtered Object
In âParameters to select cells to keepâ:
param-repeatâInsert Parameters to select cells to keepâ
âName of parameter to filter onâ: log1p_total_counts
âMin valueâ: 6.3
âMax valueâ: 20.0
Renamegalaxy-pencil output as Counts-filtered Object
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Counts-filtered Object
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: genotype
We will focus on the log1p_total_counts as that shows the biggest change. Similar to above, the bottom of the violin shape has flattered due to the threshold.
In General - Filterbycounts, you can see you now have 8,678 cells x 35,734 genes.
Hands On: Filter cells by pct_counts_mito
Scanpy FilterCells ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Counts-filtered Object
In âParameters to select cells to keepâ:
param-repeatâInsert Parameters to select cells to keepâ
âName of parameter to filter onâ: pct_counts_mito
âMin valueâ: 0
âMax valueâ: 4.5
Renamegalaxy-pencil output as Mito-filtered Object
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-filtered Object
âMethod used for plottingâ: Generic: Violin plot, using 'pl.violin'
âKeys for accessing variablesâ: Subset of variables in 'adata.var_names' or fields of '.obs'
âKeys for accessing variablesâ: log1p_total_counts,log1p_n_genes_by_counts,pct_counts_mito
âThe key of the observation grouping to considerâ: genotype
In âViolin plot attributesâ:
âAdd a stripplot on top of the violin plotâ: Yes
âAdd a jitter to the stripplotâ: Yes
âDisplay keys in multiple panelsâ: No
Renamegalaxy-pencil output Violin - Filterbymito
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Mito-filtered Object
âWhat to inspect?â: General information about the object
Renamegalaxy-pencil output General - Filterbymito
Question
Interpret the violin plot
How many genes & cells do you have in your object now?
Fantastic work! However, youâve now removed a whole heap of cells, and since the captured genes are sporadic (i.e. a small percentage of the overall transcriptome per cell) this means there are a number of genes in your matrix that are currently not in any of the remaining cells. Genes that do not appear in any cell, or even in only 1 or 2 cells, will make some analytical tools break and overall will not be biologically informative. So letâs remove them! Note that 3 is not necessarily the best number, rather it is a fairly conservative threshold. You could go as high as 10 or more.
If you are working in a group, you can now divide up a decision here with one control and the rest varied numbers so that you can compare results throughout the tutorials.
Variable: n_cells
Control > 3
Everyone else: Choose your own thresholds and compare results! Note if you go less than 3 (or even remove this step entirely), future tools are likely to fail due to empty gene data.
Hands On: Filter genes
Scanpy FilterGenes ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Mito-filtered Object
In âParameters to select genes to keepâ:
param-repeatâInsert Parameters to select genes to keepâ
âName of parameter to filter onâ: n_cells
âMin valueâ: 3
âMax valueâ: 1000000000
Renamegalaxy-pencil output as Filtered Object
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Filtered Object
âWhat to inspect?â: General information about the object
Renamegalaxy-pencil output General - Filtered object
In practice, youâll likely choose your thresholds then set up all these filters to run without checking plots in between each one. But itâs nice to see how they work!
Using the final General - Filtered object, we can summarise the results of our filtering:
Â
Cells
Genes
Raw
31178
35734
Filter genes/cell
17040
35734
Filter counts/cell
8678
35734
Filter mito/cell
8605
35734
Filter cells/gene
8605
15395
congratulations Congratulations! You have filtered your object! Now it should be a lot easier to analyse.
Processing
So currently, you have a matrix that is 8605 cells by 15395 genes. This is still quite big data. We have two issues here - firstly, you already know there are differences in how many transcripts and genes have been counted per cell. This technical variable can obscure biological differences. Secondly, we like to plot things on x/y plots, so for instance Gapdh could be on one axis, and Actin can be on another, and you plot cells on that 2-dimensional axis based on how many of each transcript they possess. While that would be fine, adding in a 3rd dimension (or, indeed, in this case, 15393 more dimensions), is a bit trickier! So our next steps are to transform our big data object into something that is easy to analyse and easy to visualise.
Hands On: Normalisation
Scanpy NormaliseData ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Filtered Object
Normalisation helps reduce the differences between gene and UMI counts by fitting total counts to 10,000 per cell. The inherent log-transform (by log(count+1)) aligns the gene expression level better with a normal distribution. This is fairly standard to prepare for any future dimensionality reduction.
Now we need to look at reducing our gene dimensions. We have loads of genes, but not all of them are different from cell to cell. For instance, housekeeping genes are defined as not changing much from cell to cell, so we could remove these from our data to simplify the dataset. We will flag genes that vary across the cells for future analysis.
Hands On: Find variable genes
Scanpy FindVariableGenes ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy NormaliseDatatool)
âFlavor of computing normalised dispersionâ: Seurat
âNumber of top variable genes to keep, mandatory if flavor=âseurat_v3ââ: `` (remove the automated 2000 here and leave the space blank)
Renamegalaxy-pencil plot output Use_me_FVG
Next up, weâre going to scale our data so that all genes have the same variance and a zero mean. This is important to set up our data for further dimensionality reduction. It also helps negate sequencing depth differences between samples, since the gene levels across the cells become comparable. Note, that the differences from scaling etc. are not the values you have at the end - i.e. if your cell has average GAPDH levels, it will not appear as a â0â when you calculate gene differences between clusters.
Hands On: Scaling data
Scanpy ScaleData ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Use_me_FVG (output of Scanpy FindVariableGenestool)
âTruncate to this value after scalingâ: 10.0
Renamegalaxy-pencil plot output Use_me_Scaled
congratulations Congratulations! You have processed your object!
Comment
At this point, we might want to remove or regress out the effects of unwanted variation on our data. A common example of this is the cell cycle, which can affect which genes are expressed and how much material is present in our cells. If youâre interested in learning how to do this, then you can move over to the Removing the Effects of the Cell Cycle tutorial now â then return here to complete your analysis.
Preparing coordinates
We still have too many dimensions. Transcript changes are not usually singular - which is to say, genes were in pathways and in groups. It would be easier to analyse our data if we could more easily group these changes.
Principal components
Principal components are calculated from highly dimensional data to find the most spread in the dataset. So in our, 3248 highly variable gene dimensions, there will be one line (axis) that yields the most spread and variation across the cells. That will be our first principal component. We can calculate the first x principal components in our data to drastically reduce the number of dimensions.
Comment: 3248???
Where did the 3248 come from? The quickest way to figure out how many highly variable genes you have, in my opinion, is to re-run galaxy-refresh the Scanpy FindVariableGenes tool and select the parameter to Remove genes not marked as highly variable. Then you can Inspect your resulting object and youâll see only 3248 genes. The following processing steps will use only the highly variable genes for their calculations, but I strongly suggest you keep even the nonvariable genes in (i.e., use the original output of your FindVariableGenes tool with way more than 3248 genes!), as a general rule. This tutorial will not work at the end plotting stage if you only take forward the 3248 or 2000 (if you set a limit on it) highly variable genes.
Warning: Check your AnnData object!
Your AnnData object should have far more than 3248 genes in it (if you followed our settings and tool versions, youâd have a matrix 8605 Ă 15395 (cells x genes). Make sure to use that AnnData object output from FindVariableGenes, rather than the 3248 or 2000 output from your testing in the section above labelled â3248â.
Hands On: Calculate Principal Components
Scanpy RunPCA ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Use_me_Scaled (output of Scanpy ScaleDatatool)
âNumber of PCs to produceâ: 50
Plot with scanpy ( Galaxy version 1.7.1+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: output_h5ad (output of Scanpy RunPCAtool)
âMethod used for plottingâ: PCA: Scatter plot in PCA coordinates, using 'pl.pca_variance_ratio'
âNumber of PCs to showâ: 50
Renamegalaxy-pencil plot output PCA Variance
Why 50 principal components you ask? Well, weâre pretty confident 50 is an over-estimate. Examine PCA Variance.
We can see that there is really not much variation explained past component 19. So we might save ourselves a great deal of time and muddied data by focusing on the top 20 PCs.
Neighborhood graph
Weâre still looking at around 20 dimensions at this point. We need to identify how similar a cell is to another cell, across every cell across these dimensions. For this, we will use the k-nearest neighbor (kNN) graph, to identify which cells are close together and which are not. The kNN graph plots connections between cells if their distance (when plotted in this 20 dimensional space!) is amonst the k-th smallest distances from that cell to other cells. This will be crucial for identifying clusters, and is necessary for plotting a UMAP. From UMAP developers: âLarger neighbor values will result in more global structure being preserved at the loss of detailed local structure. In general this parameter should often be in the range 5 to 50, with a choice of 10 to 15 being a sensible defaultâ.
If you are working in a group, you can now divide up a decision here with one control and the rest varied numbers so that you can compare results throughout the tutorials.
Control
Number of PCs to use = 20
Maximum number of neighbours used = 15
Everyone else: Use the PC variance plot to pick your own PC number, and choose your own neighbour maximum as well!b
Hands On: ComputeGraph
Scanpy ComputeGraph ( Galaxy version 1.8.1+galaxy1) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy RunPCAtool)
âUse programme defaultsâ: param-toggleNo
âMaximum number of neighbours usedâ: 15
âUse the indicated representationâ: X_pca
âNumber of PCs to useâ: 20
Dimensionality reduction for visualisation
Two major visualisations for this data are tSNE and UMAP. We must calculate the coordinates for both prior to visualisation. For tSNE, the parameter perplexity can be changed to best represent the data, while for UMAP the main change would be to change the kNN graph above itself, by changing the neighbours.
If you are working in a group, you can now divide up a decision here with one control and the rest varied numbers so that you can compare results throughout the tutorials.
Control
Perplexity = 30
Everyone else: Choose your own perplexity, between 5 and 50!
Hands On: Calculating tSNE & UMAP
Scanpy RunTSNE ( Galaxy version 1.8.1+galaxy1) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy ComputeGraphtool)
âUse the indicated representationâ: X_pca
âUse programme defaultsâ: param-toggleNo
âThe perplexity is related to the number of nearest neighbours, select a value between 5 and 50â: 30
Scanpy RunUMAP ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy RunTSNEtool)
âUse programme defaultsâ: param-toggleYes
congratulations Congratulations! You have prepared your object and created neighborhood coordinates. We can now use those to call some clusters!
Cell clusters & gene markers
Question
Letâs take a step back here. What is it, exactly, that you are trying to get from your data? What do you want to visualise, and what information do you need from your data to gain insight?
Really we need two things - firstly, we need to make sure our experiment was set up well. This is to say, our biological replicates should overlap and our variables should, ideally, show some difference. Secondly, we want insight - we want to know which cell types are in our data, which genes drive those cell types, and in this case, how they might be affected by our biological variable of growth restriction. How does this affect the developing cells, and what genes drive this? So letâs add in information about cell clusters and gene markers!
Finally, letâs identify clusters! Unfortunately, itâs not as majestic as biologists often think - the maths doesnât necessarily identify true cell clusters. Every algorithm for identifying cell clusters falls short of a biologist knowing their data, knowing what cells should be there, and proving it in the lab. Sigh. So, weâre going to make the best of it as a starting point and see what happens! We will define clusters from the kNN graph, based on how many connections cells have with one another. Roughly, this will depend on a resolution parameter for how granular you want to be.
Oh yes, yet another decision! Single cell analysis is sadly not straight forward.
Control
Resolution, high value for more and smaller clusters = 0.6
Clustering algorithm = Louvain
Everyone else: Pick your own number. If it helps, this sample should have a lot of very similar cells in it. It contains developing T-cells, so you arenât expecting massive differences between cells, like you would in, say, an entire embryo, with all sorts of unrelated cell types.
Everyone else: Consider the newer Leiden clustering method. Note that in future parameters, you will likely need to specify âleidenâ rather than âlouvainâ, which is the default, if you choose this clustering method.
Hands On: FindClusters
Scanpy FindCluster ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy RunUMAPtool)
âUse programme defaultsâ: param-toggleNo
âResolution, high value for more and smaller clustersâ: 0.6
Nearly plotting time! But one final piece is to add in SOME gene information. Letâs focus on genes driving the clusters.
FindMarkers
Hands On: FindMarkers
Scanpy FindMarkers ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: output_h5ad (output of Scanpy FindClusterstool)
But we are also interested in differences across genotype, so letâs also check that (note that in this case, itâs turning it almost into bulk RNA-seq, because youâre comparing all cells of a certain genotype against all cells of the other)
Scanpy FindMarkers ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Final object
âThe sample grouping/clustering to useâ: genotype
Do not Rename the output AnnData object (in fact, you can delete it!). You have the genotype marker table to enjoy, but we want to keep the cluster comparisons, rather than gene comparisons, stored in the AnnData object for later.
Now, thereâs a small problem here, which is that if you galaxy-eye inspect the output marker tables, you wonât see gene names, youâll see Ensembl IDs. While this is a more bioinformatically accurate way of doing this (not every ID has a gene name!), we might want to look at more well-recognised gene names, so letâs pop some of that information in!
Hands On: Adding in Gene Names
Inspect AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Final object
âWhat to inspect?â: Key-indexed annotation of variables/features (var)
This gives us our table of all the possible genes with their names.
Join two Datasets side by side on a specified field with the following parameters:
param-fileâwithâ: var (output of Inspect AnnDatatool)
âand columnâ: Column: 2
âKeep lines of first input that do not join with second inputâ: Yes
âKeep lines of first input that are incompleteâ: Yes
âFill empty columnsâ: No
âKeep the header linesâ: Yes
We have lots of extra information we donât need in our marker gene tables, soâŠ
Cut columns from a table with the following parameters:
âCut columnsâ: c1,c2,c3,c4,c11,c5,c6,c7,c8
param-fileâFromâ: param-files Select multiple files: out_file1 and output_file2 (outputs of Join two Datasetstool)
Renamegalaxy-pencil output tables Markers - cluster - named and Markers - genotype - named
congratulations Well done! Itâs time for the best bit, the plotting!
Plotting!
Itâs time! Letâs plot it all!
But first, letâs pick some marker genes from the Markers-cluster list that you made as well. Iâll be honest, in practice, youâd now be spending a lot of time looking up what each gene does (thank you google!). There are burgeoning automated-annotation tools, however, so long as you have a good reference (a well annotated dataset that youâll use as the ideal). In the mean time, letâs do this the old-fashioned way, and just copy a bunch of the markers in the original paper.
Hands On: Plot the cells!
Scanpy PlotEmbed ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Final object
âname of the embedding to plotâ: pca
âcolor by attributes, comma separated textsâ: louvain,sex,batch,genotype,Il2ra,Cd8b1,Cd8a,Cd4,Itm2a,Aif1,log1p_total_counts
âField for gene symbolsâ: Symbol
âUse raw attributes if presentâ: No
timeYou can re-run galaxy-refresh the same tool again, but change pca to tsne and then finally to umap in order to skip the following two steps.
Scanpy PlotEmbed ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Final object
âname of the embedding to plotâ: tsne
âcolor by attributes, comma separated textsâ: louvain,sex,batch,genotype,Il2ra,Cd8b1,Cd8a,Cd4,Itm2a,Aif1,log1p_total_counts
âField for gene symbolsâ: Symbol
âUse raw attributes if presentâ: No
Scanpy PlotEmbed ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Final object
âname of the embedding to plotâ: umap
âcolor by attributes, comma separated textsâ: louvain,sex,batch,genotype,Il2ra,Cd8b1,Cd8a,Cd4,Itm2a,Aif1,log1p_total_counts
âField for gene symbolsâ: Symbol
âUse raw attributes if presentâ: No
congratulations Congratulations! You now have plots galore!
Insights into the beyond
Now itâs the fun bit! We can see where genes are expressed, and start considering and interpreting the biology of it. At this point, itâs really about what information you want to get from your data - the following is only the tip of the iceberg. However, a brief exploration is good, because it may help give you ideas going forward with for your own data. Let us start interrogating our data!
Biological Interpretation
Question: Appearance is everything
Which visualisation is the most useful for getting an overview of our data, pca, tsne, or umap?
Figure 12: Louvain clustering by dimension reduction
You can see why a PCA is generally not enough to see clusters in samples - keep in mind, youâre only seeing components 1 and 2! - and therefore why the tSNE and UMAP visualisation dimensionality reductions are so useful. But there is not necessarily a clear winner between tSNE and UMAP, but I think UMAP is slightly clearer with its clusters, so weâll stick with that for the rest of the analysis.
Note that the cluster numbering is based on size alone - clusters 0 and 1 are not necessarily related, they are just the clusters containing the most cells. It would be nice to know what exactly these cells are. This analysis (googling all of the marker genes, both checking where the ones you know are as well as going through the marker tables you generated!) is a fun task for any individual experiment, so weâre going to speed past that and nab the assessment from the original paper!
The authors werenât interested in further annotation of the DP cells, so neither are we. Sometimes that just happens. The maths tries to call similar (ish) sized clusters, whether it is biologically relevant or not. Or, the question being asked doesnât really require such granularity of clusters.
If you have deviated from any of the original parameters in this tutorial, you will likely have a different number of clusters. You will, therefore, need to change the âAnnotating clustersâ âComma-separated list of new categoriesâ accordingly. Best of luck!
Annotating Clusters
Hands On: Annotating clusters
Manipulate AnnData ( Galaxy version 0.7.5+galaxy1) with the following parameters:
param-fileâAnnotated data matrixâ: Final object
âFunction to manipulate the objectâ: Rename categories of annotation
âKey for observations or variables annotationâ: louvain
âComma-separated list of new categoriesâ: DP-M4,DP-M3,DP-M1,T-mat,DN,DP-L,DP-M2,Macrophages
Hang on here, though. This unfortunately deletes the original cluster numbering. Just in case you might want this back, we can add that annotation back in.
AnnData Operations ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in hdf5 AnnData formatâ: Final object
âCopy observations (such as clusters)â: param-toggleYes
Keys from obs to copy
â+ Insert Keys from obs to copyâ
âKey containsâ: louvain
param-fileâAnnData objects with obs to copyâ: (output of Manipulate AnnDatatool)
Youâve added the new cell annotations in, now titled louvain_0. What, thatâs not good enough? You want to change the title as well? So be it.
AnnData Operations ( Galaxy version 1.8.1+galaxy0)tool with the following parameters:
param-fileâInput object in hdf5 AnnData formatâ: (output of AnnData Operationstool)
Change field names in AnnData observations
galaxy-wf-newâ+ Insert Change field names in AnnData observationsâ
1: Change field names in AnnData observations
âOriginal nameâ: louvain_0
âNew nameâ: cell_type
Renamegalaxy-pencil output h5ad Final cell annotated object
Time to re-plot! time Feel free to re-run galaxy-refresh the Scanpy PlotEmbed tool tool on the new object plotting cell_type to speed this up. OtherwiseâŠ
Scanpy PlotEmbed ( Galaxy version 1.8.1+galaxy0) with the following parameters:
param-fileâInput object in AnnData/Loom formatâ: Final cell annotated object
âname of the embedding to plotâ: umap
âcolor by attributes, comma separated textsâ: sex,batch,genotype,Il2ra,Cd8b1,Cd8a,Cd4,Itm2a,Aif1,Hba-a1,log1p_total_counts,cell_type
We can see that DP-L, which seems to be extending away from the DP-M bunch, as well as the mature T-cells (or particularly the top half) are missing some knockout cells. Perhaps there is some sort of inhibition here? INTERESTING! What next? We might look further at the transcripts present in both those populations, and perhaps also look at the genotype marker table⊠So much to investigate! But before we set you off to explore to your heartâs delight, letâs also look at this a bit more technically.
Technical Assessment
Is our analysis real? Is it right? Well, we can assess that a little bit.
While some shifts are expected and nothing to be concerned about, DP-L looks to be mainly comprised of N705. There might be a bit of batch effect, so you could consider using batch correction on this dataset. However, if we focus our attention on the other cluster - mature T-cells - where there is batch mixing, we can still assess this biologically even without batch correction.
Additionally, we will also look at the confounding effect of sex.
We note that the one female sample - unfortunately one of the mere three knockout samples - seems to be distributed in the same areas as the knockout samples at large, so luckily, this doesnât seem to be a confounding factor and we can still learn from our data. Ideally, this experiment would be re-run with either more female samples all around or swapping out this female from the male sample.
Question: Depth effect
Are there any clusters or differences being driven by sequencing depth, a technical and random factor?
Eureka! This explains the odd DP shift between wildtype and knockout cells - the left side of the DP cells simply have a higher sequencing depth (UMIs/cell) than the ones on the right side. Well, that explains some of the sub-cluster that weâre seeing in that splurge. Importantly, we donât see that the DP-L or (mostly) the mature T-cell clusters are similarly affected. So, whilst again, this variable of sequencing depth might be something to regress out somehow, it doesnât seem to be impacting our dataset. The less you can regress/modify your data, in general, the better - you want to stay as true as you can to the raw data, and only use maths to correct your data when you really need to (and not to create insights where there are none!).
Question: Sample purity
Do you think we processed these samples well enough?
We have seen in the previous images that these clusters are not very tight or distinct, so we could consider stronger filtering. Additionally, hemoglobin - a red blood cell marker that should NOT be found in T-cells - appears throughout the entire sample in low numbers. This suggests some background in the media the cells were in, and we might consider in the wet lab trying to get a purer, happier sample, or in the dry lab, techniques such as SoupX or others to remove this background. Playing with filtering settings (increasing minimum counts/cell, etc.) is often the place to start in these scenarios.
Question: Clustering resolution
Do you think the clustering is appropriate? i.e. are there single clusters that you think should be separate, and multiple clusters that could be combined?
Important to note, lest all bioinformaticians combine forces to attack the biologists: just because a cluster doesnât look like a cluster by eye is NOT enough to say itâs not a cluster! But looking at the biology here, we struggled to find marker genes to distinguish the DP population, which we know is also affected by depth of sequencing. Thatâs a reasonable argument that DP-M1, DP-M2, and DP-M3 might not be all that different. Maybe we need more depth of sequencing across all the DP cells, or to compare these explicitly to each other (consider variations on FindMarkers!). However, DP-L is both seemingly leaving the DP cluster and also has fewer knockout cells, so we might go and look at what DP-L is expressing in the marker genes. If we look at T-mat further, we can see that its marker gene - Itm2a - is only expressed in half of the cluster. You might consider sub-clustering this to investigate further, either through changing the resolution or through analysing this cluster alone.
If we look at the differences between genotypes alone (so the pseudo-bulk), we can see that most of the genes in that list are actually ribosomal. This might be a housekeeping background, this might be cell cycle related, this might be biological, or all three. You might consider investigating the cycling status of the cells, or even regressing this out (which is what the authors did).
Ultimately, there are quite a lot ways to analyse the data, both within the confines of this tutorial (the many parameters that could be changed throughout) and outside of it (batch correction, sub-clustering, cell-cycle scoring, inferred trajectories, etc.) Most analyses will still yield the same general output, though: there are fewer knockout cells in the mature T-cell population.
congratulations Congratulations! You have interpreted your plots in several important ways!
Interactive visualisations
Before we leave you to explore the unknown, you might have noticed that the above interpretations are only a few of the possible options. Plus you might have had fun trying to figure out which sample is which genotype is which sex and flicking back and forth between plots repeatedly. Figuring out which plots will be your final publishable plots takes a lot of time and testing. Luckily, there is a helpful interactive viewer Cakir et al. 2020 export tool Moreno et al. 2020 that can help you explore without having to produce new plots over and over!
Hands On: Cellxgene
Interactive CellXgene Environment with the following parameters:
param-fileâConcatenate datasetâ: Final cell annotated object
When ready, you will see a message
detailsThere is an InteractiveTool result view available, click here to display <â- Click there!
Sometimes this link can aggravate a firewall or something similar. It should be fine to go to the site. You will be asked to name your annotation, so do so to start playing around!
You can also access it by going to User in the top menu of Galaxy, then selecting Active Interactive Tools
You will need to STOP this active environment in Galaxy by going to User, Interactive Tools, selecting the environment, and selecting Stop. You may also want to delete the dataset in the history, because otherwise it continues appearing as if itâs processing.
Be warned - this visualisation tool is a powerful option for exploring your data, but it takes some time to get used to. Consider exploring it as your own tutorial for another day!
Conclusion
Hopefully, no matter which pathway of analysis you took, you found the same general interpretations. If not, this is a good time to discuss and consider with your group why that might be - what decision was âwrongâ or âill-advisedâ, and how would you go about ensuring you correctly interpreted your data in the future? Top tip - trial and error is a good idea, believe it or not, and the more ways you find the same insight, the more confident you can be! But nothing beats experimental validationâŠ
For those that did not take the âcontrolâ options, please
Rename your history (by clicking on the history title) as DECISION-Filtering and Plotting Single-cell RNA-seq Data
Add a history annotation history-annotate that includes which parameters you changed/steps you changed from the control
Sharing your history allows others to import and access the datasets, parameters, and steps of your history.
Access the history sharing menu via the History Options dropdown (galaxy-history-options), and clicking âhistory-share Share or Publishâ
Share via link
Open the History Optionsgalaxy-history-options menu at the top of your history panel and select âhistory-share Share or Publishâ
galaxy-toggleMake History accessible
A Share Link will appear that you give to others
Anybody who has this link can view and copy your history
Publish your history
galaxy-toggleMake History publicly available in Published Histories
Anybody on this Galaxy server will see your history listed under the Published Histories tab opened via the galaxy-histories-activityHistories activity
Share only with another user.
Enter an email address for the user you want to share with in the Please specify user email input below Share History with Individual Users
Your history will be shared only with this user.
Finding histories others have shared with me
Click on the galaxy-histories-activityHistories activity in the activity bar on the left
Click the Shared with me tab
Here you will see all the histories others have shared with you directly
Note: If you want to make changes to your history without affecting the shared version, make a copy by going to History Optionsgalaxy-history-options icon in your history and clicking Copy this History
In this tutorial, you moved from technical processing to biological exploration. By analysing real data - both the exciting and the messy! - you have, hopefully, experienced what itâs like to analyse and question a dataset, potentially without clear cut-offs or clear answers. If you were working in a group, you each analysed the data in different ways, and most likely found similar insights. One of the biggest problems in analysing scRNA-seq is the lack of a clearly defined pathway or parameters. You have to make the best call you can as you move through your analysis, and ultimately, when in doubt, try it multiple ways and see what happens!
We also post new tutorials / workflows there from time to time, as well as any other news.
point-right If youâd like to contribute ideas, requests or feedback as part of the wider community building single-cell and spatial resources within Galaxy, you can also join our Single cell & sPatial Omics Community of Practice.
Please also consider filling out the Feedback Form as well!
Key points
Single cell data is huge, and must have its many (# genes) dimensions reduced for analysis
Analysis is more subjective than we think, and biological understanding of the samples as well as many iterations of analysis are important to give us our best change of attaining real biological insights
Frequently Asked Questions
Have questions about this tutorial? Have a look at the available FAQ pages and support channels
Further information, including links to documentation and original publications, regarding the tools, analysis techniques and the interpretation of results described in this tutorial can be found here.
References
Trapnell, C., D. Cacchiarelli, J. Grimsby, P. Pokharel, S. Li et al., 2014 The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nature Biotechnology 32: 381â386. 10.1038/nbt.2859
Satija, R., J. A. Farrell, D. Gennert, A. F. Schier, and A. Regev, 2015 Spatial reconstruction of single-cell gene expression data. Nature Biotechnology 33: 495â502. 10.1038/nbt.3192
McCarthy, D. J., K. R. Campbell, A. T. L. Lun, and Q. F. Wills, 2017 Scater: pre-processing, quality control, normalization and visualization of single-cell RNA-seq data in R. Bioinformatics 33: 10.1093/bioinformatics/btw777
Bacon, W. A., R. S. Hamilton, Z. Yu, J. Kieckbusch, D. Hawkes et al., 2018 Single-Cell Analysis Identifies Thymic Maturation Delay in Growth-Restricted Neonatal Mice. Frontiers in Immunology 9: 10.3389/fimmu.2018.02523
Wolf, F. A., P. Angerer, and F. J. Theis, 2018 SCANPY: large-scale single-cell gene expression data analysis. Genome Biology 19: 10.1186/s13059-017-1382-0
Cakir, B., M. Prete, N. Huang, S. van Dongen, P. Pir et al., 2020 Comparison of visualization tools for single-cell RNAseq data. NAR Genomics and Bioinformatics 2: 10.1093/nargab/lqaa052
Did you use this material as an instructor? Feel free to give us feedback on how it went.
Did you use this material as a learner or student? Click the form below to leave feedback.
Hiltemann, Saskia, Rasche, Helena et al., 2023 Galaxy Training: A Powerful Framework for Teaching! PLOS Computational Biology 10.1371/journal.pcbi.1010752
Batut et al., 2018 Community-Driven Data Analysis Training for Biology Cell Systems 10.1016/j.cels.2018.05.012
@misc{single-cell-scrna-case_basic-pipeline,
author = "Wendi Bacon",
title = "Filter, plot and explore single-cell RNA-seq data with Scanpy (Galaxy Training Materials)",
year = "",
month = "",
day = "",
url = "\url{https://training.galaxyproject.org/training-material/topics/single-cell/tutorials/scrna-case_basic-pipeline/tutorial.html}",
note = "[Online; accessed TODAY]"
}
@article{Hiltemann_2023,
doi = {10.1371/journal.pcbi.1010752},
url = {https://doi.org/10.1371%2Fjournal.pcbi.1010752},
year = 2023,
month = {jan},
publisher = {Public Library of Science ({PLoS})},
volume = {19},
number = {1},
pages = {e1010752},
author = {Saskia Hiltemann and Helena Rasche and Simon Gladman and Hans-Rudolf Hotz and Delphine Larivi{\`{e}}re and Daniel Blankenberg and Pratik D. Jagtap and Thomas Wollmann and Anthony Bretaudeau and Nadia Gou{\'{e}} and Timothy J. Griffin and Coline Royaux and Yvan Le Bras and Subina Mehta and Anna Syme and Frederik Coppens and Bert Droesbeke and Nicola Soranzo and Wendi Bacon and Fotis Psomopoulos and Crist{\'{o}}bal Gallardo-Alba and John Davis and Melanie Christine Föll and Matthias Fahrner and Maria A. Doyle and Beatriz Serrano-Solano and Anne Claire Fouilloux and Peter van Heusden and Wolfgang Maier and Dave Clements and Florian Heyl and Björn GrĂŒning and B{\'{e}}r{\'{e}}nice Batut and},
editor = {Francis Ouellette},
title = {Galaxy Training: A powerful framework for teaching!},
journal = {PLoS Comput Biol}
}
Congratulations on successfully completing this tutorial!
Do you want to extend your knowledge?
Follow one of our recommended follow-up trainings:
4 stars:
Liked: Easy and very tutorial-like
Disliked: More ways to look into this data. I think that it was necessary to know this type of pipeline before, as many of this data, like the PCA, didn't tell much for me.
September 2024
4 stars:
Liked: How we were taken through the process step by step - explaining what each step achieved. This was also easy to visualise as certain steps (e.g. filtering) had been broken down into substeps so we could see the impact of each subsequent filter.
5 stars:
Liked: colours and visual display
Disliked: all good
November 2023
5 stars:
Liked: Questions and answers
September 2022
5 stars:
Liked: Clear, straight forward, but also information rich.
March 2022
5 stars:
Liked: Everything because all was new for me
Disliked: The tutorial video would be more detailed in some parts.
July 2021
2 stars:
Disliked: From find markers on I couldn't produce the desired output. If a folow the link in galaxy to Scanpy findMarkers it produces an empty table; if a seach this tool in galaxy, it produces complete tables but in the end I can't produce the plots with scanpy plot embebed (gives error). The video is to quick and does not explain what is being doing, it is just a lot of clicks. Taking in consideration the international audience, jargon should not be used. This is a shame, because it is such an interesting topic.
Wendi Bacon
Helena Rasche
Julia Jakiela
Questions:
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tabOpen image in new tab
Open image in new tab
Open image in new tabOpen image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab
Open image in new tab