##############################################
# Analysis of Longitudinal Data (SoSe 2017)  #
#                                            #
# Solution for exercise session 1 Exercise 2 #  
# Date: 04.05.2017                           #
##############################################

###############
# Exercise 2  #
###############


#### Exercise 2 a) #### 

# read rats data (eventually use setwd() to change the path)
rats <- read.csv2('rats.csv',header=T, check.names = FALSE)  # check.names = FALSE such that time variable names are not adjusted 

# transform GROUP and SUBJECT into factor variables
rats$GROUP   <- factor(rats$GROUP,labels=c('low','high','control'))
rats$SUBJECT <- factor(rats$SUBJECT)


#### Exercise 2 b) ####

# The data frame rats contains one row per subject (wide format)
# Using function melt() from package reshape2 to transform the data frame such that all measurements are in seperate rows (long format)
library(reshape2)
rats.long <- melt(rats,  id.vars = c("SUBJECT", "GROUP"), variable.name = "TIME", value.name = "RESPONSE")
# Ensure TIME is a numeric variable
rats.long$TIME <- as.numeric(as.character(rats.long$TIME))
str(rats.long)

# Instead of using the R package reshape2, you may for example also use the function reshape 
# To transform long into wide format, you might use the function dcast in reshape2

### Make plots:

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## i) Task: Explain data structure in one plot
library(ggplot2)

ggplot(rats.long, aes(x = TIME, y = RESPONSE, col = SUBJECT)) + geom_line()

## Possible solution:
# Select example curves by e.g. sampling 2 out of each group
set.seed(4332909)
sel <- numeric()
for(i in levels(rats.long$GROUP)) sel <- c(sel, sample( levels(factor(rats.long$SUBJECT[rats.long$GROUP == i])), 2) )

ggplot(rats.long, aes(x = TIME, y = RESPONSE, group = SUBJECT)) + 
  geom_line(col = "darkgrey") + 
  geom_point(col = "darkgrey") +
  geom_line(data = subset(rats.long, SUBJECT %in% sel), aes(col = GROUP), size = 2) +
  theme_bw()
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## ii) Task: Discuss data structure with researcher who might have individual informations on subjects

ggplot(rats.long, aes(x = TIME, y = RESPONSE, col = GROUP)) + geom_line() + facet_wrap(~SUBJECT)

## Possible solution
for(grp in levels(rats.long$GROUP)) {
  plot( ggplot(subset(rats.long, GROUP == grp), aes(x = TIME, y = RESPONSE, col = SUBJECT, lty = SUBJECT)) + 
      geom_line()  + 
      facet_wrap(~GROUP, labeller = label_both) +
      ylim( min(rats.long$RESPONSE, na.rm = TRUE), max(rats.long$RESPONSE, na.rm = TRUE)) )
}
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## iii) Task: Visualize group differences with group specific smooth means

ggplot(rats.long, aes(x = TIME, y = RESPONSE, col = GROUP, fill = GROUP)) + stat_smooth() + theme_bw()

## Possible solution
ggplot(rats.long, aes(x = TIME, y = RESPONSE)) + 
  geom_line(aes(group = SUBJECT), col = "darkgrey") +
  stat_smooth(aes(col = GROUP), size = 2, se = FALSE) + 
  facet_wrap(~GROUP)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  


#### Exercise 2 c) ####

# Make Lasagna Plot

## Sort with respect to average response value
# Determine order
means <- rowMeans(rats[,3:9], na.rm = TRUE)
SUBJorder <- rats[,1][order(means)]
# Adjust order
rats.long$SUBJECT <- ordered(rats.long$SUBJECT, levels = SUBJorder)

ggplot(rats.long, aes(x = TIME, y = SUBJECT, fill = RESPONSE)) + 
  geom_raster() + 
  facet_wrap(~GROUP, nrow = 3, scales = "free_y", labeller = label_both)