PreciseDist with Gephi

“An algorithm must be seen to be believed.”

-Donald Knuth

Gephi is a fabulous graphing tool, so we have made sure the results of precise_viz() can be written out into a format that Gephi understands. First, we will quickly calculate a distance matrix, transform it with precise_transform(), turn it into a graph with precise_graph(), and then write out a GraphML file with precise_viz(), which we can subsequently load into Gephi. Rather than using the output of precise_dist() with precise_fuse(), however, we will use the results of SIMLR() directly. We do this to drive home the point that precise_transform(), precise_graph() and precise_viz() are universal functions that can take any distance matrix as input. The following functions were detailed in the Cell Cycle vignette, so they are being lumped together here:

library(PreciseDist)
data("data_cell_cycle")
library(tibble)
as_tibble(data_cell_cycle)
library(dplyr)
cell_cycle_data <- data_cell_cycle %>%
  dplyr::select(-Cell_cycle) %>%
  as.matrix()
cell_cycle_labels <- data_cell_cycle %>%
  dplyr::select(Cell_cycle) %>%
  as.matrix()
library(SIMLR)
cell_cycle_simlr <- cell_cycle_data %>%
  t() %>%
  SIMLR(c = 3)$S

Now we can play with the distance matrix output of SIMLR():

cell_cycle_simlr_simil <- cell_cycle_simlr$S  %>%
  precise_transform(enforce_sim = TRUE) %>%
  precise_transform(binary_matrix = "0.20") %>%
  precise_transform(transform = "binorm") %>%
  precise_transform(transform = "random_walk") %>%
  precise_transform(enforce_dist = TRUE)

Lastly, create the graph and input it into precise_viz(). Note that we have set a file path for the graphml parameter here:

cell_cycle_graph <- precise_graph(
  data = cell_cycle_simlr_simil,
  method = 1,
  n_neighbors = 5,
  spread = 5,
  min_dist = 0.00,
  bandwidth = 1,
  parallel = FALSE,
  verbose = TRUE
)
cell_cycle_viz <- precise_viz(
  cell_cycle_graph,
  plot_type = "fr_2d_graph",
  k = 15,
  color_vec = as.character(cell_cycle_labels),
  colors = c("#0017FF", "#000000", "#FF0017"),
  size = 1,
  graphml = "/absolute_path/to_somewhere/with_full_name/not_including_the/file_extension",
  html = NULL,
  verbose = TRUE
)

The GraphML output file can now be opened in Gephi v0.9.2. To view the layout precise_viz() calculated, install the MdsLayout plugin and use gravity_x and gravity_y as the inputs into Dimension 1 and Dimension 2, respectively. Note that the colors seen upon initial loading into Gephi are written into the GraphML file after being passed from the precise_viz() color_vec and colors arguments:

Of course, you can now also leverage Gephi’s layouts like ForceAtlas2:

Or, their fruchterman-reingold layout, which gives aesthetic results very different from precise_viz():

P.S. The embedding of downloadable Gephi-exported static images into pkgdown was done using plotly_embed() from this package (i.e. PreciseDist).

Brian Muchmore

2018-09-26