Figures

Figure 1

palmerpenguins data visualized using langevitour. Interact with this figure directly or use the buttons in the text below, which will adjust the figure to demonstrate various features.

The interface allows:

• Setting a “guide” using the drop-down list. This causes to pursue projections in the vicinity of the minimum of an energy function. For example the PCA guide seeks projections with large variance in both the x and y directions. PCA None

• Hiding particular groups by unchecking their checkbox, in order to focus on other groups. For example by hiding Gentoo penguins we can focus on the difference between Adelie and Chinstrap penguins. The guide is also only applied to the visible groups. Hide Gentoo Show all

• Hiding particular axes by unchecking their checkbox. For example without bill length Adelie and Chinstrap penguins can no longer be distinguished. Hide bill length Show all

• Dragging labels on to the plot to concentrate on particular axes or try to separate out a particular group. The projection may not exactly match the label positions, since it must still be orthonormal. Bill length Bill length and depth None

• Adjusting the damping slider. High damping produces jerky Brownian motion. Less damping produces smoother less random motion. The fastest way to thoroughly explore the space of projections is an intermediate damping level. Brownian Smooth Default

• Adjusting the heat slider. More heat makes the projection move faster, and also strays further from the optimum projection defined by a guide or any labels that have been dragged on to the plot. Hot Cool Default

• Mousing over a group label. The group is highlighted.

• Mousing over an axis label. A scale and rug are displayed, and points are colored according to their position on that axis.

Figure 2

scRNA-Seq analysis using Seurat. (A) A scree plot showing variance accounted for by each PC. The scree plot has a fat tail, indicating that variation in the data can not be summarized with only a few PCs. (B) UMAP layout based on the cell scores of the first 10 PCs. U are unstimulated and S are stimulated cells.

Figure 3

scRNA-Seq cell scores visualized using langevitour.

• To simplify the dataset, hide the stimulated cells by unticking the checkbox on all the “S” labels. Show all

• To find a projection giving a good overview of the relationships between unstimulated PBMC cells, activate the “local” guide from the drop-down list.

• Examine the scale by mousing over the labels for particular axes. The scale for each component is meaningful, representing distance along a certain direction in scaled gene expression space.

• Mouse over the “U Doublet” and “S Doublet” labels to highlight doublets. Doublets located between two clusters may be interpreted as a mixture of a pair of cells with different cell types. Hiding the doublets by unchecking their checkboxes makes the clusters more distinct.

Figure 4

Noise reduction of scRNA-Seq cell scores.

Figure 5

Noise reduced scRNA-Seq cell scores.

• To provide an overview of the data, the “local” guide has already been activated.

• Doublets still lie between clusters in this denoised version. Compare this to UMAP layout, where the doublets tend to be attached to one or other cluster. To make the clusters cleaner, hide the doublets by unticking the checkboxes on the doublet labels. Show all

• To examine C1, drag the C1 label on to the plot. This pulls out the monocyte cluster. It appears monocytes are associated with C1.

• To undo this action, drag the C1 label to the right to remove it from the plot area.

Similarly C5 pulls out CD8 T cells and NK cells, and C6 pulls out B cells. The response to the cytokine is mostly contained in C3 with a further monocyte specific response in C2. Reset

Figure 6

scRNA-Seq gene loadings visualized using langevitour.

• Observe the widget spinning freely for a while. The overall geometry is of a spiky ball. The spikes generally represent sets of genes with large loadings in one component.

• Drag particular component labels onto the plot to examine them. It can be seen that biological phenomena such as different cell types (C5, C6) or response to the cytokine (C3) often involve distinct sets of genes. Furthermore, cell types are defined more by the activation of genes than the deactivation of genes. Varimax rotation was able to align these distinct sets of genes into specific components. An exception is genes used by monocytes and the monocyte response to stimulation (C1 and C2). These two components show a broad fan of genes, which can be interpreted as the genes involved in being a monocyte also being involved to varying degrees in the monocyte response to stimulation. Reset

• Mouse near points to see the specific genes they represent.