209 lines
5.8 KiB
R
209 lines
5.8 KiB
R
#!/usr/bin/env Rscript
|
|
getwd()
|
|
setwd("~/git/LSHTM_analysis/scripts/plotting/")
|
|
getwd()
|
|
#########################################################
|
|
# TASK: Basic lineage barplot showing numbers
|
|
|
|
# Output: Basic barplot with lineage samples and mut count
|
|
|
|
##########################################################
|
|
# Installing and loading required packages
|
|
##########################################################
|
|
source("Header_TT.R")
|
|
require(data.table)
|
|
source("combining_dfs_plotting.R")
|
|
# should return the following dfs, directories and variables
|
|
|
|
# PS combined:
|
|
# 1) merged_df2
|
|
# 2) merged_df2_comp
|
|
# 3) merged_df3
|
|
# 4) merged_df3_comp
|
|
|
|
# LIG combined:
|
|
# 5) merged_df2_lig
|
|
# 6) merged_df2_comp_lig
|
|
# 7) merged_df3_lig
|
|
# 8) merged_df3_comp_lig
|
|
|
|
# 9) my_df_u
|
|
# 10) my_df_u_lig
|
|
|
|
cat(paste0("Directories imported:"
|
|
, "\ndatadir:", datadir
|
|
, "\nindir:", indir
|
|
, "\noutdir:", outdir
|
|
, "\nplotdir:", plotdir))
|
|
|
|
cat(paste0("Variables imported:"
|
|
, "\ndrug:", drug
|
|
, "\ngene:", gene
|
|
, "\ngene_match:", gene_match
|
|
, "\nAngstrom symbol:", angstroms_symbol
|
|
, "\nNo. of duplicated muts:", dup_muts_nu
|
|
, "\nNA count for ORs:", na_count
|
|
, "\nNA count in df2:", na_count_df2
|
|
, "\nNA count in df3:", na_count_df3))
|
|
|
|
#===========
|
|
# input
|
|
#===========
|
|
# output of combining_dfs_plotting.R
|
|
|
|
#=======
|
|
# output
|
|
#=======
|
|
# plot 1
|
|
basic_bp_lineage = "basic_lineage_barplot.svg"
|
|
plot_basic_bp_lineage = paste0(plotdir,"/", basic_bp_lineage)
|
|
|
|
#=======================================================================
|
|
#================
|
|
# Data for plots:
|
|
# you need merged_df2, comprehensive one
|
|
# since this has one-many relationship
|
|
# i.e the same SNP can belong to multiple lineages
|
|
#================
|
|
# REASSIGNMENT as necessary
|
|
my_df = merged_df2
|
|
#my_df = merged_df2_comp
|
|
|
|
# clear excess variable
|
|
rm(merged_df2_comp, merged_df3, merged_df3_comp)
|
|
|
|
# quick checks
|
|
colnames(my_df)
|
|
str(my_df)
|
|
|
|
# Ensure correct data type in columns to plot: need to be factor
|
|
is.factor(my_df$lineage)
|
|
my_df$lineage = as.factor(my_df$lineage)
|
|
is.factor(my_df$lineage)
|
|
|
|
#==========================
|
|
# Plot: Lineage barplot
|
|
# x = lineage y = No. of samples
|
|
# col = Lineage
|
|
# fill = lineage
|
|
#============================
|
|
table(my_df$lineage)
|
|
as.data.frame(table(my_df$lineage))
|
|
|
|
#=============
|
|
# Data for plots
|
|
#=============
|
|
# REASSIGNMENT
|
|
df <- my_df
|
|
|
|
rm(my_df)
|
|
|
|
# get freq count of positions so you can subset freq<1
|
|
#setDT(df)[, lineage_count := .N, by = .(lineage)]
|
|
|
|
#******************
|
|
# generate plot: barplot of mutation by lineage
|
|
#******************
|
|
sel_lineages = c("lineage1"
|
|
, "lineage2"
|
|
, "lineage3"
|
|
, "lineage4"
|
|
#, "lineage5"
|
|
#, "lineage6"
|
|
#, "lineage7"
|
|
)
|
|
|
|
df_lin = subset(df, subset = lineage %in% sel_lineages)
|
|
|
|
# Create df with lineage inform & no. of unique mutations
|
|
# per lineage and total samples within lineage
|
|
# this is essentially barplot with two y axis
|
|
|
|
bar = bar = as.data.frame(sel_lineages) #4, 1
|
|
total_snps_u = NULL
|
|
total_samples = NULL
|
|
|
|
for (i in sel_lineages){
|
|
#print(i)
|
|
curr_total = length(unique(df$id)[df$lineage==i])
|
|
total_samples = c(total_samples, curr_total)
|
|
print(total_samples)
|
|
|
|
foo = df[df$lineage==i,]
|
|
print(paste0(i, "======="))
|
|
print(length(unique(foo$mutationinformation)))
|
|
curr_count = length(unique(foo$mutationinformation))
|
|
|
|
total_snps_u = c(total_snps_u, curr_count)
|
|
}
|
|
|
|
print(total_snps_u)
|
|
bar$num_snps_u = total_snps_u
|
|
bar$total_samples = total_samples
|
|
bar
|
|
|
|
#*****************
|
|
# generate plot: lineage barplot with two y-axis
|
|
#https://stackoverflow.com/questions/13035295/overlay-bar-graphs-in-ggplot2
|
|
#*****************
|
|
|
|
y1 = bar$num_snps_u
|
|
y2 = bar$total_samples
|
|
x = sel_lineages
|
|
|
|
to_plot = data.frame(x = x
|
|
, y1 = y1
|
|
, y2 = y2)
|
|
to_plot
|
|
|
|
# FIXME later: will be depricated!
|
|
melted = melt(to_plot, id = "x")
|
|
melted
|
|
|
|
|
|
svg(plot_basic_bp_lineage)
|
|
|
|
my_ats = 20 # axis text size
|
|
my_als = 22 # axis label size
|
|
|
|
g = ggplot(melted, aes(x = x
|
|
, y = value
|
|
, fill = variable))
|
|
|
|
printFile = g + geom_bar(stat = "identity"
|
|
, position = position_stack(reverse = TRUE)
|
|
, alpha=.75
|
|
, colour='grey75') +
|
|
theme(axis.text.x = element_text(size = my_ats)
|
|
, axis.text.y = element_text(size = my_ats
|
|
#, angle = 30
|
|
, hjust = 1
|
|
, vjust = 0)
|
|
, axis.title.x = element_text(size = my_als
|
|
, colour = 'black')
|
|
, axis.title.y = element_text(size = my_als
|
|
, colour = 'black')
|
|
, legend.position = "top"
|
|
, legend.text = element_text(size = my_als)) +
|
|
#geom_text() +
|
|
geom_label(aes(label = value)
|
|
, size = 5
|
|
, hjust = 0.5
|
|
, vjust = 0.5
|
|
, colour = 'black'
|
|
, show.legend = FALSE
|
|
#, check_overlap = TRUE
|
|
, position = position_stack(reverse = T)) +
|
|
labs(title = ''
|
|
, x = ''
|
|
, y = "Number"
|
|
, fill = 'Variable'
|
|
, colour = 'black') +
|
|
scale_fill_manual(values = c('grey50', 'gray75')
|
|
, name=''
|
|
, labels=c('Mutations', 'Total Samples')) +
|
|
scale_x_discrete(breaks = c('lineage1', 'lineage2', 'lineage3', 'lineage4')
|
|
, labels = c('Lineage 1', 'Lineage 2', 'Lineage 3', 'Lineage 4'))
|
|
|
|
print(printFile)
|
|
dev.off()
|