# Understanding Single Cell RNAseq Plots

I have just started some single cell RNAseq analysis with 500 cells. I have used Seurat for analysis. Following the guideline I have generated few plots. Now I am trying to understand how to read those plots. Can the community provide their input in the following plots. To give a brief background of the experiment. This data was processed in MinION. Its from human brain sample from our collaborator. In their shortseq data they identified 15 clusters. However, both the libraries were processed independently. The main question I am trying to answer is celltype specific isoforms. but in my dataset I can find only 6 Cell clusters. So, I am trying to understand why I am getting such a small cluster? Is it because I stared with only 500 Cells? Also I want to know what these two PCA plot is indicating. Thanks

• What are your specific questions?
– haci
Sep 9, 2019 at 16:50
• @haci am trying to understand whether the data looks good or not. This data is from long read RNAseq and I had never worked on even short read scRNA seq. Sep 9, 2019 at 16:51
• The data looking good or not is not a specific question, it is subjective. For example how many cell types or states do you expect and does it make sense (regarding your research question) to end up with 7 clusters in the UMAP plot? Unless you provide some info on your experimental set-up and research questions you cannot get useful feedback.
– haci
Sep 9, 2019 at 16:55
• @haci Thanks for your input. I am editing my question Sep 9, 2019 at 16:56
• How many PCs did you use? And how many cells are there in the dataset where 15 clusters were identified?
– haci
Sep 10, 2019 at 6:30

## 3 Answers

Identifying 15 clusters from 500 cells is challenging. It will depend on the relative abundance of each cell type within the tissue. You can use the cell type frequencies found in the shortSeq data set to see what to expect from a 500 cell dataset using the howmanycells tool

About the pc plots: The first is an elbowplot which explains how many variation is explained by each PC. The knee basically defines when components become less informative.

EDIT: I noticed that the y-axis of the elbowplot starts at 4 suggesting that the dimensionality of your data is most likely ~10 dimensions. But I expect many of the clusters that don't show up on your Umap plot to be present at very low frequencies.

The second plot is a jackstraw plot and provides a visualization of a statistical test that calculates the significance of your principal components. This one shows that all your components are significant. You can combine this with the elbowplot to estimate the dimensionality (10-15 dimensions) of your data and decide which components to use for downstream analysis. In general the more they dropoff to the dotted line the less significant they become.

While these plots are useful as a first impression, always rely on biological information and common sense to decide what information (here: components) to include in downstream analysis such as Umap, clustering, etc. For example: usually you don't want to include components mainly defined by ribosomal, mitochondrial or hsp genes.

• Whats the importance of second plot? What can I read from this plot? Sep 10, 2019 at 17:21
• @user3377241 updated my comment Sep 10, 2019 at 20:06

The first two plots are described on the Seurat PBMC tutorial. Specifically:

‘Elbow plot’: a ranking of principle components based on the percentage of variance explained by each one (ElbowPlot function).

Based on the tutorial, you should use at least 10 PCs.

The JackStrawPlot function provides a visualization tool for comparing the distribution of p-values for each PC with a uniform distribution (dashed line). ‘Significant’ PCs will show a strong enrichment of features with low p-values (solid curve above the dashed line).

Based on the tutorial, you should use more than 15 PCs.

No one can tell you the right number of clusters. If you already identified 15 clusters in a different dataset, then you should have cluster markers for those genes. You can check those genes in your dataset.

The first two plots are used in order to estimate the number of PCs to be used in later stages, for example clustering (this was split in to two functions in Seurat v3). The number of PCs selected would have an impact on the number of clusters obtained. The more PCs the more information for downstream applications. It would be a good idea to go over the genes within each PC, a trained eye can tell what kind of biology (or technical artefacts) contribute to the PCs.

Moreover, FindClusters() has a resolution parameter, which enables the number of communities (clusters) that is generated. In Seurat v3, there is a wide selection of different clustering algorithms but recommending one over the other would be beyond my expertise.

If you have a comparable dataset generated with some other technology/method, you can give a try with "integrating" the two datasets. An alternative non-Seurat approach that aims to do the same is MNNCorrect from the scran package.