Single-cell RNA sequencing is all about exploring cell subpopulations and their distinguishing properties. In this blog, you will learn how to automatically select subset of cells based on different criteria, including generating own scores. Different subsets will then be explored using previously known, as well as newly discovered marker genes.

We will work the brain cells from Drosophila melanogaster - a fruit fly, published in the study if Li et. al (2017). The data set comes with scOrange via Single Cell Datasets widget.

The data set contains more than 1800 cells labelled with different drivers, that mark different neuronal subtypes, as seen in the Distributions widget.

Here, we will focus on cells that are driven by GH146-GAL4, marking olfactory neurons - the neurons related to the sense of smell. Use Select Rows to subset the data accordingly.

Hardly ever can we escape the preprocessing step. Gene expression is quantified using counts-per-million (CPM) for each single cell, making the expression profiles comparable. Select variable genes, use a log transform and standardize expressions of each gene.

Within the olfactory projection neurons, we will try to distinguish between the lateral anterodorsal (lPN and adPN) excitatory projection neurons. The figure shows their anatomical position in relation to the antennal lobe (left) and a confocal microscopy image of GH146-GAL4+ olfactory projection neurons (Li et al., 2017).

We use Marker Genes to distinguish between the IPN and adPN lineage. Score cells will combine expression of the selected markers into an quantitative estimate of a given cell type. It was previously shown by the same group that the adPNs express the Abnormal chemosensory jump 6 transcription factor (Acj6 protein).

Using Score Cells, we can assign a score to each cell that is proportional to the expression of the acj6 gene.

Recall that the last step of preprocessing (above) performs a z-score (standardization) transformation for each gene. The values therefore represent the number of standard deviations from the gene-specific mean expression. Use Select Rows again to select cells whose marker expression (Score) is larger that 1 standard deviation. Pass the selection to t-SNE as a Data Subset to highlight them among all the cells.

We can see that the cells are well separated in the t-SNE projection, confirming the utility of acj6 in distinguishing the two neuronal subtypes.

Finally, we will investigate whether there are any genes that distinguish between two neuronal subtypes, and more interestingly, what functions are they related to. We make sure to connect the output of Select Rows to Cluster Analysis, such that all the data is passed onwards, but this time with a newly created variable Selected.

By applying Cluster Analysis on the Selected variable, we can identify the differentially expressed genes and their related functions.

Notice that acj6 is the top differentiating gene, which should be natural since our selection was based on it. Nevertheless, as observed by Li et. al 2017 (see Figure 5 therein), the adPNs (Selected = Yes) play a role in various metabolic processes, while lPNs respond to stimuli and are subject to the development of the respiratory system.

In summary, we showed how to automatically select subsets of data based on different criteria, that can also be generated dynamically.

Here is the final workflow:

References

Li, Hongjie, et al. “Classifying Drosophila olfactory projection neuron subtypes by single-cell RNA sequencing.” Cell 171.5 (2017): 1206-1220.