sankey diagram in R - data preparation

Question

I have the following data frame, where each patient is a row (I am showing only a sample of it):

df = structure(list(firstY = c("N/A", "1", "3a", "3a", "3b", "1", 
"2", "1", "5", "3b"), secondY = c("N/A", "1", "2", "3a", "4", 
"1", "N/A", "1", "5", "3b"), ThirdY = c("N/A", "1", "N/A", "3b", 
"4", "1", "N/A", "1", "N/A", "3b"), FourthY = c("N/A", "1", "N/A", 
"3a", "4", "1", "N/A", "1", "N/A", "3a"), FifthY = c("N/A", "1", 
"N/A", "2", "5", "1", "N/A", "N/A", "N/A", "3b")), class = c("tbl_df", 
"tbl", "data.frame"), row.names = c(NA, -10L))

I would like to plot a Sankey diagram, which shows the trajectory over time of each patient, and I know that I have to create nodes and links, but I'm having problems transforming the data to the format necessary to accomplish this. Specifically, the most problematic issue is to count how many patients belong to each trajectory, for example, how many patients went in the first year from stage 1 to 2, and all other combinations.

Any help with the data preparation would be appreciated.

The package Alluvial, although simple to understand, does not cope really well in case there is a lot of data.

s__ · Accepted Answer · 2019-06-05T13:35:03.753

It's not very clear what you'd like to achieve, because you do not mention the package you'd like to use, but looking at your data, it seems that this could help, if you could use the alluvial package:

library(alluvial) # sankey plots
library(dplyr)    # data manipulation

The alluvial functions can use data in wide form like yours, but it needs a frequency column, so we can create it, then do the plot:

dats_all <- df %>%                                                   # data
            group_by( firstY, secondY, ThirdY, FourthY, FifthY) %>%  # group them
            summarise(Freq = n())                                    # add frequencies

 # now plot it
alluvial( dats_all[,1:5], freq=dats_all$Freq, border=NA )

In the other hands, if you'd like to use a specific package, you should specify which.

EDIT

Using network3D is a bit tricky but you can maybe achieve some nice result from this. You need links and nodes, and have them matched, so first we can create the links:

# put your df in two columns, and preserve the ordering in many levels (columns) with paste0
links <- data.frame(source = c(paste0(df$firstY,'_1'),paste0(df$secondY,'_2'),paste0(df$ThirdY,'_3'),paste0(df$FourthY,'_4')),
                  target   = c(paste0(df$secondY,'_2'),paste0(df$ThirdY,'_3'),paste0(df$FourthY,'_4'),paste0(df$FifthY,'_5')))

# now convert as character
links$source <- as.character(links$source)
links$target<- as.character(links$target)

Now the nodes are each element in the link in a unique() way:

nodes <- data.frame(name = unique(c(links$source, links$target)))

Now we need that each nodes has a link (or vice-versa), so we match them and transform in numbers. Note the -1 at the end, because networkD3 is 0 indexes, it means that the numbers (indexes) starts from 0.

links$source <- match(links$source, nodes$name) - 1
links$target <- match(links$target, nodes$name) - 1
links$value <- 1 # add also a value

Now you should be ready to plot your sankey:

sankeyNetwork(Links = links, Nodes = nodes, Source = 'source',
              Target = 'target', Value = 'value', NodeID = 'name')

Thanks @s_t. I have a lot of data and alluvial seems very crowded. Is there a way to achive the same thing with plotly's Sankey or Network3d? — Omry Atia, Jun 05 '19 at 13:10
Firstly, thanks so much for doing this! I will check this on my full data right now — Omry Atia, Jun 05 '19 at 14:25

CJ Yetman · Answer 2 · 2021-10-23T17:21:42.957

a tidyverse way with networkd3

library(tidyr)
library(dplyr)
library(networkD3)

df <- read.table(header = TRUE, stringsAsFactors = FALSE, text = "
firstY secondY ThirdY FourthY FifthY
N/A    N/A     N/A    N/A     N/A   
1      1       1      1       1     
3a     2       N/A    N/A     N/A   
3a     3a      3b     3a      2     
3b     4       4      4       5     
1      1       1      1       1     
2      N/A     N/A    N/A     N/A   
1      1       1      1       N/A   
5      5       N/A    N/A     N/A   
3b     3b      3b     3a      3b
")

links <-
  df %>% 
  mutate(row = row_number()) %>%  # add a row id
  pivot_longer(-row, names_to = "col", values_to = "source") %>%  # gather all columns
  mutate(col = match(col, names(df))) %>%  # convert col names to col ids
  mutate(source = paste0(source, '_', col)) %>%  # add col id to node names
  group_by(row) %>%
  mutate(target = lead(source, order_by = col)) %>%  # get target from following node in row
  ungroup() %>% 
  filter(!is.na(target)) %>%  # remove links from last column in original data
  group_by(source, target) %>% 
  summarise(value = n(), .groups = "drop")  # aggregate and count similar links

# create nodes data frame from unque nodes found in links data frame
nodes <- data.frame(id = unique(c(links$source, links$target)),
                    stringsAsFactors = FALSE)
# remove column id from node names
nodes$name <- sub('_[0-9]*$', '', nodes$id)

# create node ids in links data to the 0-based index of the nodes in the nodes data frame
links$source_id <- match(links$source, nodes$id) - 1
links$target_id <- match(links$target, nodes$id) - 1

sankeyNetwork(Links = links, Nodes = nodes, Source = 'source_id',
              Target = 'target_id', Value = 'value', NodeID = 'name')

score 3 · Answer 3 · answered Jun 05 '19 at 13:34

Using ggforce:

library(ggforce)
library(dplyr)

xx <- df %>% 
  count(firstY, secondY, ThirdY, FourthY, FifthY, name = "value") %>% 
  gather_set_data(1:5) %>% 
  mutate(x = factor(x, levels = colnames(df)))


ggplot(xx, aes(x, id = id, split = y, value = value)) +
  geom_parallel_sets(alpha = 0.3, axis.width = 0.1) +
  geom_parallel_sets_axes(axis.width = 0.3) +
  geom_parallel_sets_labels(colour = "white")

sankey diagram in R - data preparation

3 Answers3

Linked