Trying out gganimate – Personal blog of Walter Muskovic

I keep putting off trying out the gganimate R package, but today’s the day. To make it more fun, rather than use the iris dataset as they’ve done in the package vignette, we’ll simulate some single cell RNA-seq data using the excellent Splatter R package.

Load R packages

library(splatter)   # simulate scRNA-seq data
library(scater)     # use the logNormCounts() function from here
library(Seurat)     # manipulating the scRNA-seq data
library(tidyverse)  # data wrangling
library(gganimate)  # fancy plots

Simulate data

I’ll simulate 400 cells transitioning along a continuous differentiation trajectory with four paths, each cell having an equal probability of belonging to each. I’ve become more familiar with the structure of Seurat objects for handling single cell data, so along the way we’ll convert the SingleCellExperiment object to a Seurat object before identifying variable features and running PCA.

# Simulate data
params.groups <- newSplatParams(batchCells = 400,
                                nGenes = 800)
sim1 <- splatSimulatePaths(params.groups,
                           path.nSteps = 10,
                           group.prob = rep(0.25, 4),
                           de.prob = 0.5,
                           de.facLoc = 0.2,
                           path.from = c(0, 1, 2, 3)) %>%
  logNormCounts() %>%
  as.Seurat() %>%
  FindVariableFeatures() %>%
  ScaleData() %>%
  RunPCA()

Peek at the metadata:

head(sim1@meta.data)
##          orig.ident nCount_originalexp nFeature_originalexp  Cell  Batch Group
## Cell1 SeuratProject              48738                  676 Cell1 Batch1 Path2
## Cell2 SeuratProject              78535                  714 Cell2 Batch1 Path2
## Cell3 SeuratProject              70102                  688 Cell3 Batch1 Path4
## Cell4 SeuratProject              73821                  700 Cell4 Batch1 Path2
## Cell5 SeuratProject              61166                  684 Cell5 Batch1 Path3
## Cell6 SeuratProject              58364                  713 Cell6 Batch1 Path1
##       ExpLibSize Step sizeFactor
## Cell1   47689.61    9  0.7940128
## Cell2   78541.88    6  1.2794491
## Cell3   71215.27    3  1.1420633
## Cell4   72928.43    2  1.2026512
## Cell5   61810.54    7  0.9964829
## Cell6   58147.56    1  0.9508343

Splatter approximates a continuous trajectory by simulating a series of steps between groups. We specified four groups and 10 steps per group. Let’s add a new variable to approximate time:

# Get group info as numeric (integer between 1-4) then multiply by ten
sim1$Time <- as.numeric(str_remove_all(sim1$Group, "Path")) * 10 
# Add step info (integer between 1-10)
sim1$Time <- sim1$Time + sim1$Step 
# start from zero not ten
sim1$Time <- sim1$Time - 10

Static plot

Create some static PCA plots highlighting the Time variable we created and the paths and steps.

# Plot group (paths), steps and our new Time variable using two of Seurat's built in 
# plot functions; DimPlot() and FeaturePlot()
DimPlot(sim1, group.by = "Group")

FeaturePlot(sim1, "Step") + scale_color_viridis_c()

FeaturePlot(sim1, "Time") + scale_color_viridis_c()

Dynamic plot

Now we’ll use the Time variable to animate the plot. We use the shadow_mark argument to keep track of the path of the cells so that they don’t just disappear from frame to frame.

# Pull data for plotting
plt.data <- cbind(sim1@meta.data,
                  Embeddings(sim1)[, 1:2])

# Create a static plot
p <- ggplot(plt.data, aes(x = PC_1, y = PC_2, col = Time)) +
  geom_point() +
  scale_colour_viridis_c()

# Add some animation
anim <- p +
  transition_states(Time,
                    transition_length = 2,
                    state_length = 1) +
  ggtitle('Time: {closest_state}') +
  shadow_mark(alpha = 0.5, size = 0.7)

animate(anim)

PCA over time with gganimate (shadow marks show prior positions).

I think the colour in the static plot does the job of conveying the trajectory of the cells through time and adding animation doesn’t contribute much. But it was good to try out gganimate anyway and discover that the grammar makes it easy to add animation to normal ggplots. Now I’ve finally tried it out I can keep an eye out for future datasets where it might come in handy. Also first post!!! 🎉 🎉