TNC_Figures_Git.Rmd

---
title: "TNC Mouse Microarray Data"
author: "AlexGW"
date: "27/09/2019"
output:
  html_document:
    df_print: paged
---

```{r setup, message=FALSE, warning=FALSE, echo=TRUE, include=TRUE}

setwd("~/Microarray_Data/bin/")

library(readxl)
library(tidyverse)
library(pheatmap)
library(gtools)
library(ggfortify)

# Load Reactome packages
library(ReactomePA)
library(org.Mm.eg.db)

# Load packages to read gmt files and use the set_colnames fxn
library(qusage)
library(magrittr)

# Load packages for limma analysis
library(biomaRt)
library(rtracklayer)
library(GenomicFeatures)
library(limma)

mouseref_8v2 <- read.csv("../input/MouseRef-8_V2_0_R3_11278551_A.csv", sep= "\t", header = T, stringsAsFactors = F)

## Need to read in raw ilmn data and work out which probe id's are to be included based on the base mean
# see bioconductor post - https://support.bioconductor.org/p/43745/
raw_limma <- read.ilmn("../input/microarray_raw_data.txt")

raw_data <- read_xlsx("../input/microarray_data_median_norm.xlsx")
mice_code <- tibble("Mice"=colnames(raw_data[2:17]),
                    "Classification"= c("Lymph_Node", 
                                        rep("Responder",5),
                                        rep("Untreated", 5), 
                                        rep("Non-Responder",5)))

mice_code[1,1] <- "DTR LN"
mice_code[3,1] <- "IT1"

# check these match as there is a slight difference in names
mice_code$Mice %>% sort()
colnames(raw_limma$E) %>% sort()

# run limma

limma_data <- neqc(raw_limma)
limma_data$other$Detection %>% colnames()

limma_order <- limma_data$other$Detection %>% colnames() %>% factor(levels = unique(.)) 
mice_code %>% arrange(match(mice_code$Mice, limma_order)) -> mice_code_arrange
##

f <- factor(mice_code_arrange$Classification)
design <- model.matrix(~0+f)
colnames(design) <- c("Lymph", "NonResponder", "Responder", "Untreated")
fit <- lmFit(limma_data, design = design)

contrast.matrix <- makeContrasts(NonResponder-Untreated, 
                                 Responder-Untreated, 
                                 NonResponder-Responder, levels = design)
fit2 <- contrasts.fit(fit, contrasts=contrast.matrix)
fit2 <- eBayes(fit2)

o <- order(fit2$Amean, decreasing=TRUE)
dup <- duplicated(fit2$genes$SYMBOL[o])
fit.unique <- fit2[o,][!dup, ]

## Done
# remove mice_code and raw_data
remove(mice_code)
remove(raw_data)

# Now we have run limma to get fit.unique first, can use this to isolate only the probe id's of highets base mean.
# read in raw data again and 
raw_data <- read_xlsx("../input/microarray_data_median_norm.xlsx") %>% 
  filter(`Probe ID`%in%(fit.unique$genes %>% rownames()))

mice_code <- tibble("Mice"=colnames(raw_data[2:17]),
                    "Classification"= c("Lymph_Node", 
                                        rep("Responder",5),
                                        rep("Untreated", 5), 
                                        rep("Non-Responder",5)))

# Create some names
unite(mice_code, full_name, 1:2) -> full_names

targets <- tibble("SampleID"=full_names$full_name, 
                  "group"=(
                    sub("^.{4}_", "", full_names$full_name)%>%{
                      sub("^.{3}_", "", (.))}%>%{
                        sub("^.{5}_", "", (.))}%>%{
                          sub("LN_Lymph_Node", "Lymph Node", (.))}
                  ))

# Show the targets file
targets

# How many are duplicated?
raw_data$ILMN_GENE %>% duplicated() %>% table()
raw_data$ILMN_GENE %>% unique() %>% length()
reformat <- raw_data[ , 1:17] %>% as_tibble() %>% `colnames<-`(c("Probe ID", full_names$full_name))

reformat <- left_join(reformat, mouseref_8v2[ , c("Probe_Id", "Entrez_Gene_ID", "Symbol", "Source")],
                      by = c("Probe ID"="Probe_Id")) %>% 
  dplyr::select(-c("Probe ID", Entrez_Gene_ID, Source) )

reformat[1:16] %>% colSums()
reformat[1:16] %>% as.matrix() %>% `rownames<-`(reformat$Symbol) -> mat_hp

```



```{r FIGURE 1 Limma analysis of DEGs, echo=T, fig.height=6, fig.width=14, message=FALSE, warning=FALSE, include=T}

# Now run the normal limma analysis on the original median normalised data (historic).
# This data was first used in other Gallimore papers and by the first author.
# Hence that was what was used for all later analyses.
# Could just use the analysis done on ILMN, but this will not match the previous figures and data.
# Also having done it results are very similar.
limma_data <- reformat[,-c(1, 17)] %>% `row.names<-`(reformat$Symbol )

f <- factor(targets[-1,]$group)
design <- model.matrix(~0+f)
colnames(design) <- c("NonResponder", "Responder", "Untreated")
fit <- lmFit(limma_data, design = design)
contrast.matrix <- makeContrasts(NonResponder-Untreated, Responder-Untreated, NonResponder-Responder, levels = design)
fit2 <- contrasts.fit(fit, contrast.matrix)
fit2 <- eBayes(fit2)

results <- decideTests(fit2)

# SAVE TWO SPREADSHEETS FOR ANA
# The Venn diagram shows the number of significant genes across the comparisons.
png(filename = "../output/vennDiagram.png", width = 24, height = 14, units = "cm", res = 300, pointsize = 8)
vennDiagram(results, mar = c(0,0,0,0), cex = c(2,2,2))
dev.off()

#Get the sig different genes and put them onto heatmaps
resp_untreat <- topTable(fit2, coef=2, adjust="BH", number = Inf) %>% 
  tibble::as_tibble(rownames="gene") %>% 
  dplyr::filter(adj.P.Val<0.05)
vect_resp_untreat <- c(2:11)

write.table(resp_untreat, file = "../output/resp_vs_untreat.txt", 
            col.names = T, row.names = F, quote = F, sep = "\t")

resp_nonresp <- topTable(fit2, coef=3, adjust="BH", number = Inf) %>% 
  tibble::as_tibble(rownames="gene") %>% 
  dplyr::filter(adj.P.Val<0.05)
vect_resp_nonresp <- c(2:6, 12:16)

write.table(resp_nonresp, file = "../output/resp_vs_nonresp.txt", 
            col.names = T, row.names = F, quote = F, sep = "\t")

plot_hp <- function(genes, vector, nom){ mat_hp[rownames(mat_hp)%in%genes, vector] %>%  {
  pheatmap((.), scale = "none", 
           cutree_cols = 1,
           cluster_cols = F,
           cluster_rows = T,
           treeheight_col = 20,
           gaps_col = 5,
           silent = F, 
           show_colnames = F,
           show_rownames = T, 
           clustering_method = "ward.D2", 
           clustering_distance_rows = "manhattan",
           clustering_distance_cols = "manhattan",
           color = colorRampPalette(
             c("navy","white", "firebrick3"))((50)),
           main = nom,
           fontsize = 14)}
}

pathway_fxn <- function(gene){
  mouseref_8v2[ (mouseref_8v2$Symbol %in% gene), "Entrez_Gene_ID"] %>% 
    base::unique() %>% {
      enrichPathway((.), organism = "mouse",
                    pvalueCutoff = 0.05, 
                    pAdjustMethod = "BH", # Benjamini Hochberg
                    qvalueCutoff = 1,
                    #universe = universeEntrez,
                    minGSSize = 4, 
                    maxGSSize = Inf, readable = T) } -> overall
  barplot(overall, showCategory=12, 
          cex.axis = 2, cex.names = 2, 
          xlab = "Number of genes belonging to pathway", axes = T,
          axisnames = T)
}

# Non-Responders High
graphics.off()
png(filename = "../output/nonresp_high_hp.png", height = 50*0.2, width = 15, units = "cm", res = 300, pointsize = 1)
plot_hp(resp_nonresp %>% dplyr::filter(logFC>0) %>% pull(gene) %>% 
          as.character(), vect_resp_nonresp, "")
dev.off()

# What pathways are higher in the non responders?
#pathway_fxn(resp_nonresp %>% dplyr::filter(logFC>0) %>% pull(gene) %>% as.character())

# Responders High
graphics.off()
png(filename = "../output/resp_high_hp.png", height = 50*0.74, width = 15, units = "cm", res = 300, pointsize = 1)
plot_hp(resp_nonresp %>% dplyr::filter(logFC<0) %>% pull(gene) %>% 
          as.character(), vect_resp_nonresp, " ")
dev.off()

# SUPP FIGURE
graphics.off()
png(filename = "../output/resp_high_pathways.png", width = 30, height = 10, units = "cm", res = 300, pointsize = 12)
pathway_fxn(resp_nonresp %>% dplyr::filter(logFC<0) %>% pull(gene) %>% as.character())
dev.off()

# Expand non responders to p < 0.10 - SUPP FIG
resp_nonresp_relax <- topTable(fit2, coef=3, adjust="BH", number = Inf) %>% 
  tibble::as_tibble(rownames="gene") %>% 
  dplyr::filter(adj.P.Val<0.10)

graphics.off()
png(filename = "../output/nonresp_high_p010_hp.png", width = 10, height = 46, units = "cm", res = 300, pointsize = 1)
plot_hp(resp_nonresp_relax %>% dplyr::filter(logFC>0) %>% pull(gene) %>% as.character(), vect_resp_nonresp, "")
dev.off()

graphics.off()
png(filename = "../output/nonresp_high_p010_pathways.png", width = 30, height = 5, units = "cm", res = 300, pointsize = 12)
pathway_fxn(resp_nonresp_relax %>% dplyr::filter(logFC>0) %>% pull(gene) %>% as.character())
dev.off()

```



```{r FIGURE 1 C New heatmap, echo=T, fig.height=6, fig.width=14, message=FALSE, warning=FALSE, include=T}

annot <- data.frame(#"LogFoldChange"=resp_nonresp$logFC, "Padj"=resp_nonresp$adj.P.Val, 
  row.names = resp_nonresp$gene,
  "Expression"= factor(resp_nonresp$logFC<0) %>% recode_factor(`FALSE` = "Non Responder High",
                                                               `TRUE` = "Responder High"))

plot_hp <- function(genes, vector, nom){ mat_hp[rownames(mat_hp)%in%genes, vector] %>% 
    t %>%  { 
      # not sure about the reason for the rownames of rownames inclusion
      pheatmap((.), scale = "none", 
               cutree_cols = 2,
               cluster_cols = T,
               treeheight_col = 20,
               
               annotation_col = annot,
               annotation_names_col = F,
               annotation_names_row = F,
               annotation_legend = F,
               
               gaps_row = 5,
               cutree_rows = 1,
               cluster_rows = F,
               silent = F, 
               show_colnames = T,
               show_rownames = F, 
               clustering_method = "ward.D2", 
               clustering_distance_rows = "manhattan",
               clustering_distance_cols = "manhattan",
               color = colorRampPalette(
                 c("navy","white", "firebrick3"))((50)),
               main = nom,
               fontsize = 16)}
}

#New Fig. 1C - This made it to the paper
graphics.off()
png(filename = "../output/new_hp_format.png", height = 12.5, width = 50, units = "cm", res = 300, pointsize = 1)
plot_hp(resp_nonresp %>%  pull(gene) %>% 
          as.character(), vect_resp_nonresp, "")
dev.off()


```


```{r Manhattan Distances, include=T, echo=T, message=FALSE, warning=FALSE, fig.height=6, fig.width=14}

run_hp <- function(Genes, Name) {
  go <- mat_hp[(rownames(mat_hp)) %in% (Genes), vect_resp_nonresp] %>% t
  
  go  %>%  {
    pheatmap((.), scale = "none",
             cluster_cols = T,
             cluster_rows = F,
             gaps_row = c(5),
             cutree_cols = 4,
             treeheight_col = 20,
             treeheight_row = 20,
             silent = F,
             show_colnames = T,
             show_rownames = T,
             clustering_method = "ward.D2",
             clustering_distance_rows = "manhattan",
             clustering_distance_cols = "manhattan",
             color = colorRampPalette(c("navy", "white", "firebrick3"))((50)),
             main = Name,
             fontsize = 8)}
}

# We need to find what Tnc is clustering with.
library(parallel)
library(parallelDist)
library(reshape2)
library(magrittr)
options(digits=4)

# Work out manhattan distance of tnc with all other genes. Use responders and non responders only.

dist4 <- parDist(mat_hp[ , vect_resp_nonresp], method = 'manhattan', threads = 8)

#dist4 <- parDist((mat_hp[, -1 ]), method = 'manhattan', threads = 6)

# turn into a matrix, reorganise into var1, var2 and then dist value
# subset out Tnc, sort the distance values of Tnc from everything and draw heatmap of top 100 Tnc associated genes

m1 <- as.matrix(dist4)
m1[upper.tri(m1)] <- 999
#m2 <- m1
m2 <- subset(melt(m1), value!=999)

# this is a function that outputs the desired heatmaps and csv files from a distance analysis
manhattan <- function(gene){
  goi <- deparse(substitute(gene))
  
  m2 <- m2[ m2$Var2==goi, ]
  m2 <- m2[order(m2$value), 3:1]
  write.table(m2, file = paste0("../output/", goi, "dist_col.txt"), row.names=F, sep="\t", quote=F)
  
  dist_ls <- m2 %>% arrange(desc(value))
  
  closest <- as.character( dist_ls[(nrow(dist_ls)-29):nrow(dist_ls), 3])
  furthest <- as.character( dist_ls[1:30, 3])
  
  x <- dist_ls[(nrow(dist_ls)-29):nrow(dist_ls), c(1, 3) ] %>% set_colnames(c("Manhattan Distance", "Gene")) %>% 
    arrange(`Manhattan Distance`) %>% mutate(`Manhattan Distance`=round(`Manhattan Distance`, digits = 2))
  
  write.table(x, file = paste0("../output/", goi, "closest30.csv"), 
              row.names = F, col.names = T, sep = ",", quote = F)
  
  graphics.off()
  png(filename = paste0("../output/", goi, "closest30.png"), 
      width = 18, height = 8, units = "cm", res = 300, pointsize = 1)
  run_hp(c(closest, goi), 
         paste0("Bottom 30 Genes with Expression Patterns that Match ", goi))
  dev.off()
  
  
  x <- dist_ls[1:30, c(1, 3)]  %>% set_colnames(c("Manhattan Distance", "Gene")) %>% 
    arrange( desc(`Manhattan Distance`)) %>% mutate(`Manhattan Distance`=round(`Manhattan Distance`, digits = 2)) 
  
  write.table(x, file = paste0("../output/", goi, "furthest30.csv"), 
              row.names = F, col.names = T, sep = ",", quote = F)
  
  graphics.off()
  png(filename = paste0("../output/", goi, "furthest30.png"), 
      width = 18, height = 8, units = "cm", res = 300, pointsize = 1)
  run_hp(c(furthest, goi), 
         paste0("Top 30 Genes with Expression Patterns Opposite to ", goi))
  dev.off()
}

manhattan(Tnc)

# could have tried a spearman or pearson correlation instead.

```

\newline
This section was to run lots of PCAs and look at pathways which give a high standard deviation and separate out the groups well. Didn't make it into the final publication. Also here we had to make sure the genes corresponded correctly between mouse and humans, hence the use of biomart. so the relevant code has been left in.
\newline

```{r Pathway Setup, message=FALSE, warning=FALSE, echo=TRUE, include=TRUE, results="hide"}

# Reactome Pathway PCA Analysis
all_paths_gmt <- read.gmt("../resources/ReactomePathways.gmt") %>% map(function(x){ x[-1]} ) # Remove the "Reactome pathways"
gmt_df <- tibble("Pathway"=map(all_paths_gmt, unlist)%>%names(), "Genes"=map(all_paths_gmt, unlist) )

# read in all reactome paths
all_paths <- read.table("../resources/ReactomePathways.txt", sep = "\t", header = F) %>% 
  as_tibble() %>%
  filter({grepl("MMU", (.)$V1)}) %>% 
  filter({grepl("Mus musculus", (.)$V3)}) %>% 
  set_colnames(c("Code", "Pathway", "Species") ) %>%
  left_join(gmt_df, by="Pathway") %>%
  # Use dplyr here!!!
  add_column("Number"=map_dbl((.)$Genes, length)) %>% 
  dplyr::arrange(desc(Number)) #%>%
#filter(Number>5) #%>%
#filter(Number<1000) # Get rid of large pathways with over 1000 genes

genes <-  all_paths$Genes %>% unlist() %>% unique()

# Basic function to convert human to mouse gene names
human <- useEnsembl("ensembl", dataset = "hsapiens_gene_ensembl")
mouse <- useEnsembl("ensembl", dataset = "mmusculus_gene_ensembl")
# genesV2 <- getLDS(attributes = c("hgnc_symbol"), # Attributes you want to retrieve of primary dataset. 
#                   # A possible list of attributes can be retrieved using the function listAttributes.
#                   filters = "hgnc_symbol", # Filters that should be used in the query. 
#                   values = genes, 
#                   mart = human, 
#                   attributesL = c("mgi_symbol"), # Attributes of linked dataset that needs to be retrieved
#                   martL = mouse, # Mart object representing linked dataset
#                   uniqueRows=T)

# Here is one I made earlier
#saveRDS(genesV2, file = "../output/genesV2.Rdata", compress = F)
genesV2 <- readRDS(file = "../output/genesV2.Rdata")

# name the character vector correctly
genes_recode <- as.character(genesV2$MGI.symbol) %>% `names<-`(genesV2$HGNC.symbol)

# library(future)
# library(furrr)
# # What I am doing below is so hard that you need a big PC and then use parallel mapping
# plan(multiprocess)

# What this is doing below, is going through the all_paths$Genes area...
# And if it finds a human gene name (which will be in capitals) in the genes_recode named character vector above...
# Then it renames (recodes) it as the mouse equivalent...
# It is not as simple as just changing to small letters...

# These ways may interact with the bioMart server, but that can sometimes fail...
# THAT is why I am doing it this way.

# get_res <- furrr::future_map(.x=all_paths$Genes, 
#                              .f = possibly(function(x) {
#                                recode(.x=x, !!!genes_recode,
#                                       # above you use a named character vector and
#                                       # to unquote it with !!! (splicing)
#                                       # this whole area of quos and non standard evaluation...
#                                       # Tough
#                                       default="no_match", .missing="missing")}, 
#                                otherwise = "error") )

# Here is one I made earlier
# saveRDS(get_res, file = "../output/get_res.Rdata", compress = F)
get_res <- readRDS("../output/get_res.Rdata")

# Now this taking the gene names in all_paths and converting them to mouse...
# Again, you can specify to look at mouse genes in the reactome code...

#       enrichPathway((.), organism = "mouse",

# However, what i did here was to put all the genes from pathways onto a PCA..
# And this had to be done myself, hence the gene name change...
all_paths_convert <- all_paths %>% 
  mutate("Genes"=get_res) %>% mutate("Genes"=map((.)$Genes, unique)) %>% 
  mutate("Number"=map_dbl((.)$Genes, length))

```

\newline
Create heartbeat plots of gene expression associated with ECM and adaptive immune pathways.
\newline

```{r Heartbeat Plots adaptive Imm and ECM organisation, include=T, echo=T, message=FALSE, warning=FALSE, results='asis', fig.height=8, fig.width=24}

library(ggplot2)

vect_resp_nonresp <- c(2:6, 12:16)

genes1 <- "Adaptive Immune System"
genes2 <- "Extracellular matrix organization"

library(forcats) # for working with factors

out_genes <- tibble( "Genes"= all_paths_convert[grepl(genes1, all_paths_convert$Pathway), ] %>% 
                       pull( "Genes") %>% unlist(), "Category"=genes1 )
out_genes_2 <- tibble( "Genes"=all_paths_convert[grepl(genes2, all_paths_convert$Pathway), ] %>% 
                         pull( "Genes") %>% unlist(), "Category"=genes2 )
out_genes_df <- bind_rows(out_genes, out_genes_2)

dup_ids <- out_genes_df %>% filter(duplicated(Genes))

# so wierd the way this is the way to remove genes
out_genes_df[ out_genes_df$Genes %in% dup_ids$Genes, "Category"] <- "Both Pathways"

remove(out_genes, out_genes_2)

out_genes_df <- out_genes_df %>% filter(Genes %>% duplicated() %>% !.)

# FIND THE GENES WITH HIGHEST DIFFERENCE BETWEEN MEANS
# This is for the plots - so there are only a sElection of genes labelled
# Work out the mean difference between genes in responders and non-responders

mean_diff <- mat_hp[rownames(mat_hp) %in% (out_genes_df%>%pull(Genes)), vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df, by="Genes") %>%
  # mutate(Genes=(Genes)) %>% # what was this done for?
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  add_column(R_mean={rowMeans(.[, c(2:6)])}) %>% add_column(NR_mean={rowMeans(.[, c(7:11)])}) %>%
  add_column(Abs_Mean_Diff=abs(.$R_mean - .$NR_mean)) %>% filter(Abs_Mean_Diff>1.00) 

###
# it gets used later

library(ggpubr)

mat_hp[rownames(mat_hp) %in% (out_genes_df%>%filter(Category=="Adaptive Immune System")%>%pull(Genes)), 
       vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df%>%filter(Category=="Adaptive Immune System"), by="Genes") %>%
  #mutate(Genes=(Genes)) %>% # Why did you do this?
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  gather(2:11, key = Dataset, value = Expression) %>%
  add_column("Response"=sub(.$Dataset, pattern = "^(.*?)_", replacement = "")) %>%
  #filter(Category=="Adaptive Immune System") %>%
  {ggplot((.), aes(x=Genes, y=Expression, fill=Response, color=Response) ) + 
      geom_point(position = position_dodge(width = 1), alpha=0.8, size=1.5) +
      #facet_wrap(~Category, ncol = 1, scales = "free") + 
      theme_pubr() + 
      theme(legend.position = "bottom") +
      theme(axis.text.x = element_text(angle = -90, vjust = 0.5, hjust = 0, size = 18),
            panel.grid.major.x = element_line(color = "black"),
            text = element_text(size=18) ) +
      stat_summary(fun.y=mean, geom="line", aes(group=Response), size=2, alpha=0.99) +
      ylab("Normalised Gene Expression") + xlab(element_blank()) +
      scale_x_discrete(breaks = (.)$Genes[(.)$Genes %in% mean_diff$Genes]) # then use mean diff here
    #ggtitle("Adaptive Immune System Genes")
  }

ggsave(filename = "../output/AIS_heartbeat.png", plot = last_plot(), device = "png", dpi = 300, width = 18, height =6)

mat_hp[rownames(mat_hp) %in% (out_genes_df%>%filter(Category=="Extracellular matrix organization")%>%pull(Genes)), 
       vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df%>%filter(Category=="Extracellular matrix organization"), by="Genes") %>%
  #mutate(Genes=(Genes)) %>% 
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  gather(2:11, key = Dataset, value = Expression) %>%
  add_column("Response"=sub(.$Dataset, pattern = "^(.*?)_", replacement = "")) %>%
  #  filter(Category=="Extracellular matrix organization") %>%
  {ggplot((.), aes(x=Genes, y=Expression, fill=Response, color=Response) ) + 
      geom_point(position = position_dodge(width = 1), alpha=0.8, size=1.5) +
      #facet_wrap(~Category, ncol = 1, scales = "free") + 
      theme_pubr() + 
      theme(legend.position = "bottom") +
      theme(axis.text.x = element_text(angle = -90, vjust = 0.5, hjust = 0, size = 18),
            panel.grid.major.x = element_line(color = "black"),
            text = element_text(size=18) ) +
      stat_summary(fun.y=mean, geom="line", aes(group=Response), size=2, alpha=0.99) +
      ylab("Normalised Gene Expression") + xlab(element_blank()) +
      scale_x_discrete(breaks = (.)$Genes[(.)$Genes %in% mean_diff$Genes])
    #ggtitle("Extracellular Matrix Organisation Genes")
  }

ggsave(filename = "../output/ECM_heartbeat.png", plot = last_plot(), device = "png", dpi = 300, width = 18, height =6)

```


```{r Heartbeat Plots Mean Difference, include=T, echo=T, message=FALSE, warning=FALSE, results='asis', fig.height=6, fig.width=24}

mean_diff <- mat_hp[rownames(mat_hp) %in% (out_genes_df%>%pull(Genes)), vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df, by="Genes") %>%
  # mutate(Genes=(Genes)) %>% # what was this done for?
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  add_column(R_mean={rowMeans(.[, c(2:6)])}) %>% add_column(NR_mean={rowMeans(.[, c(7:11)])}) %>%
  add_column(Abs_Mean_Diff=abs(.$R_mean - .$NR_mean)) %>% filter(Abs_Mean_Diff>0.50) 

annot <- mean_diff[, "Category"] %>%  set_rownames(mean_diff$Genes) %>% 
  as.data.frame(stringsAsFactors=T)

run_hp <- function(Genes, Name) {
  
  go <- mat_hp[(rownames(mat_hp)) %in% (Genes), c(2:6, 12:16)] %>% t
  
  go  %>%  {
    pheatmap((.), scale = "none",
             cluster_cols = T,
             cluster_rows = F,
             annotation_col = annot,
             annotation_colors = list(`Category`=c(`Adaptive Immune System`="dodgerblue1",
                                                   `Both Pathways`="dimgray",
                                                   `Extracellular matrix organization`="gold1")),
             annotation_names_col = T,
             annotation_names_row = F,
             annotation_legend = T,
             treeheight_col = 20,
             cutree_cols = 2,
             treeheight_row = 20,
             silent = F,
             gaps_row = 5,
             show_colnames = F,
             show_rownames = F,
             clustering_method = "ward.D2",
             clustering_distance_rows = "manhattan",
             clustering_distance_cols = "manhattan",
             color = colorRampPalette(
               c("navy","white", "firebrick3"))((50)),
             #main = Name,
             fontsize = 8)}
}

graphics.off()
png(filename = "../output/pathway_comparison.png", width = 24, height = 8, units = "cm", res = 300, pointsize = 12)
run_hp(Genes = mean_diff$Genes, "")
dev.off()

### ECM Heatmap
mean_diff <- mat_hp[rownames(mat_hp) %in% (out_genes_df%>%pull(Genes)), vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df, by="Genes") %>%
  # mutate(Genes=(Genes)) %>% # what was this done for?
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  add_column(R_mean={rowMeans(.[, c(2:6)])}) %>% add_column(NR_mean={rowMeans(.[, c(7:11)])}) %>%
  add_column(Abs_Mean_Diff=abs(.$R_mean - .$NR_mean)) %>% filter(Abs_Mean_Diff>0.50) %>%
  filter(Category=="Extracellular matrix organization")

annot <- mean_diff[, "Category"] %>%  set_rownames(mean_diff$Genes) %>% 
  as.data.frame(stringsAsFactors=T)

run_hp <- function(Genes, Name) {
  
  go <- mat_hp[(rownames(mat_hp)) %in% (Genes), c(2:6, 12:16)] %>% t
  
  go  %>%  {
    pheatmap((.), scale = "none",
             cluster_cols = T,
             cluster_rows = F,
             annotation_col = annot,
             annotation_colors = list(`Category`=c(#`Adaptive Immune System`="dodgerblue1",
               #`Both Pathways`="dimgray",
               `Extracellular matrix organization`="gold1")),
             annotation_legend = F,
             annotation_names_row = F,
             annotation_names_col = F,
             treeheight_col = 20,
             cutree_cols = 2,
             treeheight_row = 20,
             silent = F,
             gaps_row = 5,
             show_colnames = T,
             show_rownames = F,
             clustering_method = "ward.D2",
             clustering_distance_rows = "manhattan",
             clustering_distance_cols = "manhattan",
             color = colorRampPalette(
               c("navy","white", "firebrick3"))((50)),
             #main = Name,
             fontsize = 14)} 
}

graphics.off()
png(filename = "../output/ecm_hp.png", width = 30, height = 10, units = "cm", res = 300, pointsize = 12)
run_hp(Genes = mean_diff$Genes, "")
dev.off()

### Adaptive Immune Heatmap

mean_diff <- mat_hp[rownames(mat_hp) %in% (out_genes_df%>%pull(Genes)), vect_resp_nonresp] %>% 
  as_tibble(rownames="Genes") %>% 
  left_join(out_genes_df, by="Genes") %>%
  # mutate(Genes=(Genes)) %>% # what was this done for?
  arrange(Category) %>% mutate(Genes=factor(Genes, levels = .$Genes) ) %>%
  add_column(R_mean={rowMeans(.[, c(2:6)])}) %>% add_column(NR_mean={rowMeans(.[, c(7:11)])}) %>%
  add_column(Abs_Mean_Diff=abs(.$R_mean - .$NR_mean)) %>% filter(Abs_Mean_Diff>0.50) %>%
  filter(Category=="Adaptive Immune System")

annot <- mean_diff[, "Category"] %>%  set_rownames(mean_diff$Genes) %>% 
  as.data.frame(stringsAsFactors=T)

run_hp <- function(Genes, Name) {
  
  go <- mat_hp[(rownames(mat_hp)) %in% (Genes), c(2:6, 12:16)] %>% t
  go  %>%  {
    pheatmap((.), scale = "none",
             cluster_cols = T,
             cluster_rows = F,
             annotation_col = annot,
             annotation_colors = list(`Category`=c(`Adaptive Immune System`="dodgerblue1"
                                                   #`Both Pathways`="dimgray",
                                                   #`Extracellular matrix organization`="gold1"
             )),
             annotation_legend = F,
             annotation_names_row = F,
             annotation_names_col = F,
             treeheight_col = 20,
             cutree_cols = 2,
             treeheight_row = 20,
             silent = F,
             gaps_row = 5,
             show_colnames = T,
             show_rownames = F,
             clustering_method = "ward.D2",
             clustering_distance_rows = "manhattan",
             clustering_distance_cols = "manhattan",
             color = colorRampPalette(
               c("navy","white", "firebrick3"))((50)),
             #main = Name,
             fontsize = 14)}
}

graphics.off()
png(filename = "../output/ais_hp.png", width = 40, height = 10, units = "cm", res = 300, pointsize = 12)
run_hp(Genes = mean_diff$Genes, "")
dev.off()

```


\newline
Analyse cancer stem cell genes. Gene list was decided by Ana and Awen.
\newline


```{r CSC Genes, fig.width=8, fig.height=6, message=FALSE, warning=FALSE, echo=TRUE, include=TRUE, results="asis"}

csc_genes_human <- c("Zeb1", "Zeb2", "Twist1", "Twist2", "Snai1", "CD24A", "THY1", "MCAM", "TTC3", "VLDLR") %>% toupper()

# Basic function to convert human to mouse gene names
genesCSC <- getLDS(attributes = c("hgnc_symbol"), # Attributes you want to retrieve of primary dataset. 
                   # A possible list of attributes can be retrieved using the function listAttributes.
                   filters = "hgnc_symbol", # Filters that should be used in the query. 
                   values = csc_genes_human, 
                   mart = human, 
                   attributesL = c("mgi_symbol"), # Attributes of linked dataset that needs to be retrieved
                   martL = mouse, # Mart object representing linked dataset
                   uniqueRows=T)

csc_genes <- genesCSC$MGI.symbol

run_hp <- function(genes, nom){
  out_genes <- genes
  go <- mat_hp[(rownames(mat_hp)) %in% (out_genes), vect_resp_nonresp%>%rev() ] 
  
  go  %>% t() %>% {
    pheatmap((.), scale = "none",
             cluster_cols = T,
             cluster_rows = F,
             gaps_row = 5,
             silent = F,
             show_colnames = T,
             show_rownames = F,
             clustering_method = "ward.D2",
             clustering_distance_rows = "manhattan",
             clustering_distance_cols = "manhattan",
             
             color = colorRampPalette(
               c("navy","white", "firebrick3"))((50)),
             fontsize = 14)}
}
graphics.off()
png(filename = "../output/csc_hp.png", width = 10, height = 10, units = "cm", res = 300, pointsize = 1)
run_hp(csc_genes, "Cancer Stem Cell Genes")
dev.off()

# GENES
mat_hp[rownames(mat_hp) %in% csc_genes, 
       vect_resp_nonresp %>% rev() ] %>% 
  as_tibble(rownames="Genes") %>%  gather(2:11, key = Dataset, value = Expression) %>%
  add_column("Response"=sub(.$Dataset, pattern = "^(.*?)_", replacement = "")) %>%
  #filter(Category=="Adaptive Immune System") %>%
  {ggplot((.), aes(x=Genes, y=Expression, fill=Response, color=Response) ) + 
      geom_point(position = position_dodge(width = 0.25), alpha=0.8, size=1.5) +
      #facet_wrap(~Category, ncol = 1, scales = "free") + 
      scale_color_manual(values = c("#F8766D", "#00BFC4")%>%rev() ) +
      scale_fill_manual(values = c("#F8766D", "#00BFC4")%>%rev()) +  
      theme_pubr() + 
      theme(legend.position = "bottom",
            legend.text = element_text(size = 18),
            axis.text.x = element_text(angle = -90, vjust = 0.5, hjust = 0, size = 18),
            panel.grid.major.x = element_line(color = "grey90"),
            text = element_text(size=18) ) +
      stat_summary(fun.y=mean, geom="line", aes(group=Response), size=2, alpha=0.99) +
      ylab("Normalised Gene Expression") + xlab(element_blank())
  }

ggsave(filename = "../output/CSC_heartbeat.png", plot = last_plot(), device = "png", dpi = 300, width = 6, height =6)

```