TCGA_Analysis_2019_cluster_TNC_First.Rmd

---
title: "TNC & TCGA Analysis"
output:
  html_document:
    df_print: paged
---

```{r setup, message=FALSE, warning=FALSE, results="hide"}

setwd("~/TNC/2019_TNC_Reanalysis/bin")

library(survival)
library(survminer)
library(rjson)
library(tidyverse)
library(magrittr)
library(ClusterR)
library(data.table)

```


```{r json processing}

# https://docs.gdc.cancer.gov/Encyclopedia/pages/TCGA_Barcode/

clinical_json <- fromJSON(file = "../input/clinical.cart.2019-10-22.json", simplify = F)

clinical <- map_chr(.x = 1:length(clinical_json), 
                    .f= function(x){clinical_json[[x]]$case_id}) %>% 
  tibble("case_id"=.) %>% as_tibble() %>%
  
  add_column("submitter_id"=  map(.x = 1:length(clinical_json), 
                                  .f= possibly(
                                    .f =function(x){clinical_json[[x]]$diagnoses[[1]]$submitter_id}, 
                                    otherwise = NULL) )) %>%
  mutate(submitter_id=sub(pattern = "_diagnosis", replacement =  "", submitter_id)) %>%
  
  add_column("vital_status"=map(.x = 1:length(clinical_json), 
                                .f= possibly(
                                  function(x){clinical_json[[x]]$demographic$vital_status}, 
                                  otherwise = NULL))%>% as.character() ) %>%
  
  add_column("days_to_death"=map(.x = 1:length(clinical_json), 
                                 .f= possibly(
                                   function(x){clinical_json[[x]]$demographic$days_to_death}, 
                                   otherwise = NULL)) ) %>%
  mutate(days_to_death=as.integer(days_to_death %>% as.character())) %>%
  
  add_column("days_to_last_follow_up"=map(.x = 1:length(clinical_json), 
                                          .f= possibly(
                                            function(x){clinical_json[[x]]$diagnoses[[1]]$days_to_last_follow_up}, 
                                            otherwise = NULL)) ) %>%
  mutate(days_to_last_follow_up=as.integer(days_to_last_follow_up %>% as.character())) %>%
  
  add_column("tumor_stage"=map(.x = 1:length(clinical_json), 
                               .f= possibly(
                                 function(x){clinical_json[[x]]$diagnoses[[1]]$tumor_stage}, 
                                 otherwise = NULL))%>% as.character() ) %>%
  
  add_column("tumor_grade"=map(.x = 1:length(clinical_json), 
                               .f= possibly(
                                 function(x){clinical_json[[x]]$diagnoses[[1]]$tumor_grade}, 
                                 otherwise = NULL))%>% as.character() )

```


```{r create meta}

sample_sheet <- read.table("../input/gdc_sample_sheet.2019-10-22.tsv", sep = "\t",
                           header = T, stringsAsFactors = F ) %>% as_tibble()

sample_clinical_df <- left_join(sample_sheet %>% 
                                  mutate(File.Name=sub(pattern = ".htseq.counts.gz", 
                                                       replacement = "",  sample_sheet$File.Name)), 
                                clinical, by = c("Case.ID"="submitter_id"), keep=T)

files <-  sub(pattern = ".gz", replacement = "",  sample_sheet$File.Name)

```


```{r read file}

library(data.table)

TCGA_data <- map_dfc(.x= paste0("../TCGA_3_10_2019/files/", files) , .f= fread, drop = 1, data.table = FALSE,
                     nThread = 6,
                     header = FALSE, colClasses = c("numeric") )

# gene_check <- map_dfc(.x= paste0("../TCGA_3_10_2019/files/", files)[c(1:100, 9000:9500)] , 
#                       .f= fread, drop = 2, data.table = FALSE,
#                             nThread = 6,
#                             header = FALSE, colClasses = c("character") )
# 
# map_chr(1:600, function(i){ identical(gene_check[[1]], gene_check[[i]])}) %>% table
# Nice.

use_4_name <- read.table("../TCGA_3_10_2019/files/0000772b-773d-4cf8-8baf-0e1e6dbf55e8.htseq.counts",
                         header = F, col.names = c("gene", "count"))


colnames(TCGA_data) <- basename(sub(".htseq.counts", "", files) )
rownames(TCGA_data) <- substring(use_4_name$gene, 1, 15)

complete_df <- TCGA_data %>% t() %>% as_tibble(rownames = "File.Name") %>% 
  inner_join(sample_clinical_df, by = "File.Name")

```
\newline

\newline
```{r deseq2, message=FALSE, warning=FALSE, results="hide"}

library(DESeq2)
library(biomaRt)
library(rtracklayer)
library(GenomicFeatures)
library(parallel)
library(BiocParallel)

matrix_TCGA_data <- as.matrix(TCGA_data[1:(nrow(TCGA_data)-5), ]) # Remove the last 5 rows

exptDesign_TCGA <- data.frame(
  row.names = colnames(matrix_TCGA_data),
  condition = complete_df$Project.ID)
# gene_check <- map_dfc(.x= paste0("../TCGA_3_10_2019/files/", files)[c(1:100, 9000:9500)] , 
#                       .f= fread, drop = 2, data.table = FALSE,
#                             nThread = 6,
#                             header = FALSE, colClasses = c("character") )
# 
# map_chr(1:600, function(i){ identical(gene_check[[1]], gene_check[[i]])}) %>% table

dds <- DESeqDataSetFromMatrix(
  countData = matrix_TCGA_data,
  colData = exptDesign_TCGA,
  design = ~ 1) 

dds <- DESeq2::estimateSizeFactors(dds)

norm_counts <- DESeq2::counts(dds, normalized=TRUE)

norm_clinical_df <- norm_counts %>% t() %>% as_tibble(rownames = "File.Name") %>% 
  inner_join(sample_clinical_df, by = "File.Name")

```

```{r TPM}
# This was not used here - the TPM counts were used for Immune deconvolution that did not make the final manuscript.

# Make TPM counts
# https://support.bioconductor.org/p/91218/

# https://gdc.cancer.gov/about-data/data-harmonization-and-generation/gdc-reference-files
# https://docs.gdc.cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/

# get gene lengths there
# look @ Rob's code
# txdb <- makeTxDbFromGRanges(import("../resources/gencode.v22.annotation.gtf"))
# allTranscripts <- transcriptLengths(txdb)
# allGeneIDs <- unique(allTranscripts$gene_id)
# gene_len_df <- allTranscripts %>% 
#   group_by(gene_id) %>% summarize(max.tx_len = max(tx_len)) %>%
#   as_tibble %>% mutate(gene_id=map_chr(gene_id, substring, 1, 15))

# Is this the same as the raw data?
# counts.mat <- counts(dds, normalized=FALSE)
# Yes - other than one is integer and one numeric
# identical(counts.mat%>%as.numeric, 
#           matrix_TCGA_data%>%as.numeric)

# Are the genes the same?
# identical(counts.mat%>%rownames, 
#           gene_len_df$gene_id)
# Yeah

# gene.length <- gene_len_df$max.tx_len
# x <- counts.mat / gene.length
# tpm.mat <- t( t(x) * 1e6 / colSums(x) )
# 
# tpm_clinical_df <- tpm.mat %>% t() %>% as_tibble(rownames = "File.Name") %>% 
#   inner_join(sample_clinical_df, by = "File.Name")

# check TPM clinical

```
\newline

\newline
```{r gene signatures}

TNC_signature <- c('TNC')
TNC_genes <- c('ENSG00000041982')

CTL_signature <- c('CD3D',	'CD3E',	'CD3G',	'CD8A',	'PTPRC')
CTL_genes <- c('ENSG00000167286',	'ENSG00000198851',	'ENSG00000160654',	
               'ENSG00000153563',	'ENSG00000081237')

CytoTox_signature <- c('GNLY',	'GZMA',	'GZMB',
                       'GZMH',	'GZMK',	'PRF1')
CytoTox_genes <- c('ENSG00000115523',	'ENSG00000145649',	'ENSG00000100453',
                   'ENSG00000100450',	'ENSG00000113088',	
                   'ENSG00000180644')

CSC_signature <- c('VLDLR', 'TTC3', 'MCAM', 'TWIST1', 'ZEB1', 
                   'SNAI1', 'TWIST2', 'SNAI3', 'THY1', 'ZEB2', 
                   'CD24A', 'PROM1')

CSC_genes <- c('ENSG00000147852', 'ENSG00000182670', 'ENSG00000076706', 'ENSG00000122691', 'ENSG00000148516',
               'ENSG00000124216', 'ENSG00000233608', 'ENSG00000185669', 'ENSG00000154096', 'ENSG00000169554',
               'ENSG00000272398', 'ENSG00000007062')

signature <- c(TNC_signature,
               CTL_signature, 
               CytoTox_signature,
               CSC_signature)

genes <- c(TNC_genes, 
           CTL_genes, 
           CytoTox_genes,
           CSC_genes)

# check this with biomart
mart <- useMart(biomart = "ensembl",dataset="hsapiens_gene_ensembl")
all_names <- getBM(filters= "ensembl_gene_id",
                   attributes= c("external_gene_name"),
                   values= genes,
                   mart= mart)

df <- norm_clinical_df %>% dplyr::select( c("Project.ID",  "Case.ID", "Sample.ID",  
                                            "Sample.Type", "File.Name",
                                            "vital_status", 
                                            "days_to_death",  "days_to_last_follow_up",
                                            #"tumor_stage", # you aren't using tumor stage here
                                            all_of(genes)) ) %>% # why use all_of?
  dplyr::filter(Sample.Type=="Primary Tumor") %>% as_tibble()

colnames(df)[9:(8+length(genes))] <- signature

# Are there multipe samples for the same biopsy 
# df$Sample.ID %>% duplicated %>% table
# df$Case.ID %>% duplicated %>% table

# df %>% filter( duplicated(Case.ID)|duplicated(Sample.ID) ) %>% 
#   pull(Case.ID) %>% {filter(.data = df, Case.ID %in% (.))} %>% 
#   filter(!duplicated(Case.ID)) %>%
#   filter(!duplicated(Sample.ID)) %>%
#   arrange(Sample.ID) %>%
#   arrange(Case.ID)
# Yes
# https://www.biostars.org/p/311017/
# https://www.biostars.org/p/310587/
# https://www.biostars.org/p/308192/

# These duplicates must be removed
# nrow(df)

df <- df %>%  filter(!duplicated(Case.ID)) %>%
  filter(!duplicated(Sample.ID))

nrow(norm_clinical_df %>% dplyr::select( c("Project.ID",  "Case.ID", "Sample.ID",  
                                           "Sample.Type", "File.Name") ) %>% 
       dplyr::filter(Sample.Type=="Primary Tumor"))

# nrow(df)

```
\newline

Look at the expression of CTL and TNC - this is why we keep the different cancers separate.\newline

\newline
```{r Gene Expression Plots, fig.height=6, fig.width=12}

df %>% gather(key = "Gene", value = "Counts", all_of( c(CTL_signature, CytoTox_signature))) %>% 
  ggplot(aes(x= Project.ID, y= log(Counts), color = Project.ID)) +
  #ggplot(aes(x= interaction(Gene, Project.ID), y= log(Counts), color = Project.ID)) +
  geom_point(position = "jitter", size = 0.5) + theme_bw() + theme(legend.position = "none") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log CTL Expression") + xlab(NULL) +
  #facet_wrap(~Gene, ncol = 4)
  # ylim( 0, 150000) + 
  ggsave("../output/NU_CTL_Norm_exp_cancers.png",  device = "png", units = "cm", plot = last_plot(),
         width = 20, height = 10, dpi = 300)

df %>% gather(key = "Gene", value = "Counts", TNC) %>% 
  ggplot(aes(x= Project.ID, y= log(Counts), color = Project.ID)) +
  geom_point(position = "jitter", size = 0.5) + theme_bw() + theme(legend.position = "none") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log TNC Expression") + xlab(NULL) +
  ggsave("../output/NU_TNC_Norm_exp_cancers.png",  device = "png", units = "cm", plot = last_plot(),
         width = 20, height = 10, dpi = 300)

# based on this graph, BCRA, LUAD and SARC are the cancers where TNC expression varies most
# maybe COAD and PAAD

```

```{r survival setup}

df_survival <- df %>% as_tibble %>% 
  add_column("times"=NA %>% as.integer(), .before=8)

for (i in 1:length(df_survival[[1]])){
  if (is.na( (df_survival[i , ]$days_to_death)) ){
    df_survival[i , ]$times <- df_survival[i , ]$"days_to_last_follow_up"
  }else
  {
    df_survival[i , ]$times <- df_survival[i , ]$"days_to_death"
  }
}

df_survival$times <- as.integer(df_survival$times)
# df_survival$vital_status %>% table
# df_survival$vital_status %>% is.na %>% table
# 6 are not reported.
# 5 are NA
df_survival$vital_status <- recode(df_survival$vital_status, 
                                   Alive="0", Dead="1", `Not Reported` = NA_character_
)
# df_survival$vital_status %>% table
# df_survival$vital_status %>% is.na %>% table

# check this, where does it say 0 and 1 
# in the survminer vignetter, not survival vignette
# patient.vital_status = survival status. 0 = alive, 1 = dead
df_survival$vital_status <- as.numeric(df_survival$vital_status)

comp_analysis <- df_survival[ 
  !(is.na( df_survival$times))&!(is.na( df_survival$vital_status)), 
  ] %>%  as_tibble 

# No of patients with no times or vital status
# nrow(df_survival)-nrow(comp_analysis)

```
\newline
2.1 Analysis in the paper CTL First, TNC after.\newline
\newline
```{r Standard Clustering Analysis}

cancers_tb_removed <- c("TCGA-KIRC", "TCGA-KIRP", "TCGA-LGG", "TCGA-PCPG", 
                        "TCGA-THCA", "TCGA-PRAD", "TCGA-PAAD")

full_data_creation <- function(to_be_removed = "NULL"){
  
  data_for_cluster <- comp_analysis  %>% filter( !(Project.ID %in% to_be_removed))
  
  tcell <- c(CytoTox_signature, CTL_signature)
  
  project_ids <- data_for_cluster$Project.ID %>% table() %>% names()
  
  cluster_by_ctl <- map(.x= project_ids, .f= function(x){  
    
    go <- data_for_cluster %>% filter(Project.ID %in% x)
    
    medoids_tnc <- Cluster_Medoids( center_scale( as.matrix(go %>% dplyr::select(all_of(tcell))), 
                                                  mean_center = T, sd_scale = T ),
                                    clusters = 2, 
                                    distance_metric = "manhattan", 
                                    swap_phase = T, 
                                    seed = 123)
    
    go <- go %>% mutate("cluster"= (medoids_tnc$clusters %>% as.character))
    
    clust_1 <- map(go %>% filter(cluster=="1") %>% # why not try map_dbl
                     dplyr::select(all_of(tcell)), .f = mean) %>% # get the mean of each tcell gene
      unlist %>% mean # get the mean of gene means
    
    clust_2 <- map(go %>% filter(cluster=="2") %>% 
                     dplyr::select(all_of(tcell)), .f = mean) %>% 
      unlist %>% mean
    
    go <- go %>% mutate(cluster= case_when(
      (clust_1 > clust_2)&(.$cluster=="1")  ~ "CTL High",
      (clust_1 > clust_2)&(.$cluster=="2")  ~ "CTL Low",
      (clust_2 > clust_1)&(.$cluster=="2")  ~ "CTL High",
      (clust_2 > clust_1)&(.$cluster=="1")  ~ "CTL Low")) %>%
      as.data.frame
    
    return(go)
  })
  names(cluster_by_ctl) <- project_ids
  
  ## Second Level
  cluster_ctl_by_tnc <- map(.x= cluster_by_ctl, .f= function(x){
    
    go <- x
    
    medoids_ctl_high <- Cluster_Medoids( center_scale( 
      as.matrix(go %>% filter(cluster=="CTL High") %>% dplyr::select("TNC")), 
      mean_center = T, sd_scale = T ),
      clusters = 2, 
      distance_metric = "manhattan", 
      swap_phase = T, 
      seed = 123)
    
    medoids_ctl_low <- Cluster_Medoids( center_scale( 
      as.matrix(go %>% filter(cluster=="CTL Low") %>% dplyr::select("TNC")), 
      mean_center = T, sd_scale = T ),
      clusters = 2, 
      distance_metric = "manhattan", 
      swap_phase = T, 
      seed = 123)
    
    ### CTL High
    go[ go$cluster=="CTL High", "cluster2"] <- medoids_ctl_high$clusters %>% as.character
    ### CTL Low
    go[ go$cluster=="CTL Low", "cluster2"] <- medoids_ctl_low$clusters %>% as.character
    
    mean_fxn <- function(clus1, clus2){
      map(go %>% filter(cluster==!!clus1) %>%
            filter(cluster2==!!clus2) %>% 
            dplyr::select("TNC"), .f = mean) %>%  # for each group get the column mean
        unlist }
    
    clust_ctlhi_1 <- mean_fxn("CTL High", "1")
    clust_ctlhi_2 <- mean_fxn("CTL High", "2")
    clust_ctllo_1 <- mean_fxn("CTL Low", "1")
    clust_ctllo_2 <- mean_fxn("CTL Low", "2")
    
    go <- go %>% mutate(cluster2= case_when(
      (.$cluster=="CTL High")&(clust_ctlhi_1 > clust_ctlhi_2)&(.$cluster2=="1") ~ "TNC High",
      (.$cluster=="CTL High")&(clust_ctlhi_1 > clust_ctlhi_2)&(.$cluster2=="2") ~ "TNC Low",
      (.$cluster=="CTL High")&(clust_ctlhi_2 > clust_ctlhi_1)&(.$cluster2=="2") ~ "TNC High",
      (.$cluster=="CTL High")&(clust_ctlhi_2 > clust_ctlhi_1)&(.$cluster2=="1") ~ "TNC Low",
      (.$cluster=="CTL Low")&(clust_ctllo_1 > clust_ctllo_2)&(.$cluster2=="1")  ~ "TNC High",
      (.$cluster=="CTL Low")&(clust_ctllo_1 > clust_ctllo_2)&(.$cluster2=="2")  ~ "TNC Low",
      (.$cluster=="CTL Low")&(clust_ctllo_2 > clust_ctllo_1)&(.$cluster2=="2")  ~ "TNC High",
      (.$cluster=="CTL Low")&(clust_ctllo_2 > clust_ctllo_1)&(.$cluster2=="1")  ~ "TNC Low"
    )) %>% as.data.frame
    
    return(go)  
    
  })
  
  return(cluster_ctl_by_tnc)
}

all_cancer <- full_data_creation()
ctl_corr_cancers <- full_data_creation(cancers_tb_removed)

```



```{r figure 2, fig.height=18, fig.width=12}
# does TNC correlate with survival?
go <- map_df(ctl_corr_cancers, ~as.data.frame(.))

fit <- survfit(Surv( times, vital_status ) ~ cluster,
               data = go )

# FIGURE 2A
survp <- ggsurvplot(fit, go,
                    legend.labs = c("CTL High", "CTL Low"),
                    palette = c("darkorange1", "seagreen4"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("CTL Clustering"),
                    xlab = c("Time (days)"),
                    tables.y.text = F,
                    ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                      font.x = c(12, "bold.italic", "darkred"),
                                      font.y = c(12, "bold.italic", "darkred"),
                                      font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Fig_2A.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

# FIGURE 2B
fit <- survfit(Surv( times, vital_status ) ~ cluster2,
               data = go[ go$cluster=="CTL High", ])
survp <- ggsurvplot(fit, go[ go$cluster=="CTL High", ],
                    legend.labs = c("TNC High", "TNC Low"),
                    palette = c("firebrick3", "dodgerblue3"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("CTL High & TNC Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    ncol=2)  %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Fig_2B.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

# FIGURE 2C
fit <- survfit(Surv( times, vital_status ) ~ cluster2,
               data = go[ go$cluster=="CTL Low", ])
survp <- ggsurvplot(fit, go[ go$cluster=="CTL Low", ],
                    legend.labs = c("TNC High", "TNC Low"),
                    palette = c("firebrick3", "dodgerblue3"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("CTL Low & TNC Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                      font.x = c(12, "bold.italic", "darkred"),
                                      font.y = c(12, "bold.italic", "darkred"),
                                      font.legend = c(12, "plain", "black"))

ggsave(filename =  paste("../output/", "Fig_2C.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

```

```{r Supp Fig 4, fig.height=10, fig.width=12}

# SUPP FIGURE 4A
go <- map_df(all_cancer, ~as.data.frame(.))

fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster,
               data = go )

survp <- ggsurvplot(fit, go,
                    legend.labs = c("CTL High", "CTL Low"),
                    palette = c("darkorange1", "seagreen4"),
                    facet.by = "Project.ID", 
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("CTL Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    
                    ncol=3, scales = "fixed") %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                                        font.x = c(12, "bold.italic", "darkred"),
                                                        font.y = c(12, "bold.italic", "darkred"),
                                                        font.legend = c(12, "plain", "black"))

ggsave(filename =  paste("../output/", "Supp_Fig3.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40, plot = print( survp))
survp

# Have a look at the results of clustering here

go %>% gather(all_of( c(CytoTox_signature, CTL_signature)), key = "Tcell", value = "Exp") %>%
  ggplot(aes(x = Project.ID, y = Exp, fill = cluster)) + 
  geom_point(position = position_dodge(width = 0.75), size = 0.75) + theme_bw() + 
  theme(legend.position = "bottom") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log Gene Expression") + xlab(NULL) + 
  #scale_y_continuous(limits = c(0, 4e+05)) +
  facet_wrap(ncol = 2, ~Tcell, scales = "free") + 
  ggsave("../output/NU_gene_exp_clusters_tcell.png", 
         dpi = 300, device = "png", width = 26, height = 32)

go %>% ggplot(aes(x = Project.ID, y = TNC, fill = cluster2)) + 
  geom_point(position = position_dodge(width = 0.75), size = 0.75) + theme_bw() + 
  theme(legend.position = "bottom") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log TNC Expression") + xlab(NULL) + 
  scale_y_continuous(limits = c(0, 4e+05)) +
  facet_wrap(ncol = 1, ~cluster, scales = "free") + 
  ggsave("../output/NU_gene_exp_clusters_tnc.png", 
         dpi = 300, device = "png", width = 14, height = 8)

# Other way

go %>% gather(all_of( c(CytoTox_signature, CTL_signature)), key = "Tcell", value = "Exp") %>%
  ggplot(aes(x = Project.ID, y = Exp, fill = cluster2)) + 
  geom_point(position = position_dodge(width = 0.75), size = 0.75) + theme_bw() + 
  theme(legend.position = "bottom") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log Gene Expression") + xlab(NULL) + 
  #scale_y_continuous(limits = c(0, 4e+05)) +
  facet_wrap(ncol = 2, ~Tcell, scales = "free") + 
  ggsave("../output/NU_gene_exp_clusters_tcell_in_tnc.png", 
         dpi = 300, device = "png", width = 26, height = 32)

go %>% ggplot(aes(x = Project.ID, y = TNC, fill = cluster)) + 
  geom_point(position = position_dodge(width = 0.75), size = 0.75) + theme_bw() + 
  theme(legend.position = "bottom") +
  geom_boxplot(alpha=0.4) +
  theme(text = element_text(size = 10),
        plot.title = element_text(size = 10),
        axis.text.y = element_text(size = 10),
        axis.text.x = element_text(angle = 270, size = 10),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank()
  ) + ylab("Log TNC Expression") + xlab(NULL) + 
  scale_y_continuous(limits = c(0, 4e+05)) +
  facet_wrap(ncol = 1, ~cluster2, scales = "free") + 
  ggsave("../output/NU_gene_exp_clusters_tnc_in_tcell.png", 
         dpi = 300, device = "png", width = 14, height = 8)

```

```{r write tpm csv}

# colnames(go)
# colnames(tpm_clinical_df[1])
# 
# joined <- left_join(go, tpm_clinical_df %>% dplyr::filter(Sample.Type=="Primary Tumor"), 
#                     by = c("File.Name"))
# 
# fwrite(x= joined, file = "../output/denconv_clusters.csv",
#        sep = ",", row.names = F, col.names = T)

```


```{r Supp Fig 4B, fig.height=14, fig.width=8}

go <- map_df(ctl_corr_cancers, ~as.data.frame(.))

#SUPP  FIGURE 4A
fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster2,
               data = go[ go$cluster=="CTL High", ])

survp <-  ggsurvplot(fit = fit, data = go[ go$cluster=="CTL High", ], 
                     legend.labs = c("TNC High", "TNC Low"),
                     palette = c("firebrick3", "dodgerblue3"),
                     risk.table = T, 
                     risk.table.height = 0.3,
                     pval = TRUE, 
                     pval.coord = c(6000, 0.9),
                     ggtheme = theme_survminer(),
                     #xlim=c(0,5000), #
                     #break.x.by = 2000,
                     title = c("CTL High & TNC Clustering"),
                     font.main = c(12, "bold", "darkblue"),
                     font.x = c(12, "bold.italic", "darkred"),
                     font.y = c(12, "bold.italic", "darkred"),
                     font.tickslab = c(12, "plain", "darkgreen"),
                     xlab = c("Time (days)"),
                     tables.y.text =F,
                     
                     facet.by = "Project.ID",
                     ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Supp_Fig4.png", sep = ""),
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40,
       plot = print( survp))
survp

```

```{r Supp Fig 4C, fig.height=14, fig.width=8}

# SUPP FIGURE 5
fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster2,
               data = go[ go$cluster=="CTL Low", ])
survp <-  ggsurvplot(fit = fit, data = go[ go$cluster=="CTL Low", ], 
                     risk.table = T, 
                     risk.table.height = 0.3,
                     legend.labs = c("TNC High", "TNC Low"),
                     palette = c("firebrick3", "dodgerblue3"),
                     pval = TRUE, 
                     pval.coord = c(6000, 0.9),
                     ggtheme = theme_survminer(),
                     #xlim=c(0,5000), #
                     #break.x.by = 2000,
                     title = c("CTL Low & TNC Clustering"),
                     font.main = c(12, "bold", "darkblue"),
                     font.x = c(12, "bold.italic", "darkred"),
                     font.y = c(12, "bold.italic", "darkred"),
                     font.tickslab = c(12, "plain", "darkgreen"),
                     xlab = c("Time (days)"),
                     tables.y.text =F,
                     
                     facet.by = "Project.ID",
                     ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Supp_Fig5.png", sep = ""),
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40,
       plot = print( survp))
survp

```
\newline

\newline
```{r reverse analysis}

reverse_data_creation <- function(to_be_removed = "NULL"){
  
  data_for_cluster <- comp_analysis  %>% filter( !(Project.ID %in% to_be_removed))
  
  tcell <- c(CytoTox_signature, CTL_signature)
  
  project_ids <- data_for_cluster$Project.ID %>% table() %>% names()
  
  cluster_by_tnc <- map(.x= project_ids, .f= function(x){  
    
    go <- data_for_cluster %>% filter(Project.ID %in% x)
    
    medoids_tnc <- Cluster_Medoids( center_scale( as.matrix(go %>% dplyr::select("TNC")), 
                                                  mean_center = T, sd_scale = T ),
                                    clusters = 2, 
                                    distance_metric = "manhattan", 
                                    swap_phase = T, 
                                    seed = 123)
    
    go <- go %>% mutate("cluster"= (medoids_tnc$clusters %>% as.character))
    
    clust_1 <- map(go %>% filter(cluster=="1") %>% 
                     dplyr::select("TNC"), .f = mean) %>% unlist
    clust_2 <- map(go %>% filter(cluster=="2") %>% 
                     dplyr::select("TNC"), .f = mean) %>% unlist
    
    go <- go %>% mutate(cluster= case_when(
      (clust_1 > clust_2)&(.$cluster=="1")  ~ "TNC High",
      (clust_1 > clust_2)&(.$cluster=="2")  ~ "TNC Low",
      (clust_2 > clust_1)&(.$cluster=="2")  ~ "TNC High",
      (clust_2 > clust_1)&(.$cluster=="1")  ~ "TNC Low")) %>%
      as.data.frame
    
    return(go)
  })
  names(cluster_by_tnc) <- project_ids
  
  ## Second Level
  cluster_tnc_by_ctl <- map(.x= cluster_by_tnc, .f= function(x){
    
    go <- x
    
    medoids_tnc_high <- Cluster_Medoids( center_scale( 
      as.matrix(go %>% filter(cluster=="TNC High") %>% dplyr::select(all_of(tcell))), 
      mean_center = T, sd_scale = T ),
      clusters = 2, 
      distance_metric = "manhattan", 
      swap_phase = T, 
      seed = 123)
    
    medoids_tnc_low <- Cluster_Medoids( center_scale( 
      as.matrix(go %>% filter(cluster=="TNC Low") %>% dplyr::select(all_of(tcell))), 
      mean_center = T, sd_scale = T ),
      clusters = 2, 
      distance_metric = "manhattan", 
      swap_phase = T, 
      seed = 123)
    
    ### TNC High
    go[ go$cluster=="TNC High", "cluster2"] <- medoids_tnc_high$clusters %>% as.character
    ### TNC Low
    go[ go$cluster=="TNC Low", "cluster2"] <- medoids_tnc_low$clusters %>% as.character
    
    mean_fxn <- function(clus1, clus2){
      map(go %>% filter(cluster==!!clus1) %>%
            filter(cluster2==!!clus2) %>% 
            dplyr::select(all_of(tcell)), .f = mean) %>% # get the mean of each column
        unlist %>% mean} # get the mean of all the col means
    
    clust_hi_1 <- mean_fxn("TNC High", "1")
    clust_hi_2 <- mean_fxn("TNC High", "2")
    clust_lo_1 <- mean_fxn("TNC Low", "1")
    clust_lo_2 <- mean_fxn("TNC Low", "2")
    
    go <- go %>% mutate(cluster2= case_when(
      (.$cluster=="TNC High")&(clust_hi_1 > clust_hi_2)&(.$cluster2=="1") ~ "CTL High",
      (.$cluster=="TNC High")&(clust_hi_1 > clust_hi_2)&(.$cluster2=="2") ~ "CTL Low",
      (.$cluster=="TNC High")&(clust_hi_2 > clust_hi_1)&(.$cluster2=="2") ~ "CTL High",
      (.$cluster=="TNC High")&(clust_hi_2 > clust_hi_1)&(.$cluster2=="1") ~ "CTL Low",
      (.$cluster=="TNC Low")&(clust_lo_1 > clust_lo_2)&(.$cluster2=="1")  ~ "CTL High",
      (.$cluster=="TNC Low")&(clust_lo_1 > clust_lo_2)&(.$cluster2=="2")  ~ "CTL Low",
      (.$cluster=="TNC Low")&(clust_lo_2 > clust_lo_1)&(.$cluster2=="2")  ~ "CTL High",
      (.$cluster=="TNC Low")&(clust_lo_2 > clust_lo_1)&(.$cluster2=="1")  ~ "CTL Low"
    )) %>% as.data.frame
    
    return(go)  
    
  })
  
  return(cluster_tnc_by_ctl)
}

rev_analysis <- reverse_data_creation(cancers_tb_removed)

```

```{r Supp Figure 6, fig.height=4, fig.width=6}
go_rev <- map_df(rev_analysis, ~as.data.frame(.))

# SUPP FIGURE 6A
fit <- survfit(Surv( times, vital_status ) ~ cluster,
               data = go_rev )
survp <- ggsurvplot(fit, go_rev,
                    legend.labs = c("TNC High", "TNC Low"),
                    palette = c("firebrick3", "dodgerblue3"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("TNC Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    ncol=2)  %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Supp_Fig6A.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

# FIGURE 6B
fit <- survfit(Surv( times, vital_status ) ~ cluster2,
               data = go_rev[ go_rev$cluster=="TNC High", ])

survp <- ggsurvplot(fit, go_rev[ go_rev$cluster=="TNC High", ],
                    legend.labs = c("CTL High", "CTL Low"),
                    palette = c("darkorange1", "seagreen4"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("TNC High & CTL Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    ncol=2)  %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Supp_Fig6B.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

# SUPP FIGURE 6C
fit <- survfit(Surv( times, vital_status ) ~ cluster2,
               data = go_rev[ go_rev$cluster=="TNC Low", ])
survp <- ggsurvplot(fit, go_rev[ go_rev$cluster=="TNC Low", ],
                    legend.labs = c("CTL High", "CTL Low"),
                    palette = c("darkorange1", "seagreen4"),
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("TNC Low & CTL Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    ncol=2)  %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))
ggsave(filename =  paste("../output/", "Supp_Fig6C.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 14, height = 12, plot = print( survp))
survp

```


```{r Unused Reverse Facet Plots, fig.height=26.5, fig.width=12}

# TNC Survival
fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster,
               data = go_rev )
survp <- ggsurvplot(fit, go_rev,
                    legend.labs = c("TNC High", "TNC Low"),
                    palette = c("firebrick3", "dodgerblue3"),
                    facet.by = "Project.ID", 
                    risk.table = T, 
                    risk.table.height = 0.3,
                    pval = TRUE, 
                    pval.coord = c(6000, 0.9),
                    ggtheme = theme_survminer(),
                    #xlim=c(0,5000), 
                    break.x.by = 2000,
                    title = c("TNC Clustering"),
                    font.main = c(12, "bold", "darkblue"),
                    font.x = c(12, "bold.italic", "darkred"),
                    font.y = c(12, "bold.italic", "darkred"),
                    font.tickslab = c(12, "plain", "darkgreen"),
                    xlab = c("Time (days)"),
                    tables.y.text =F,
                    
                    ncol=2)  %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))

ggsave(filename =  paste("../output/", "NU_Reverse_TNC_cluster.png", sep = ""), 
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40, plot = print( survp))


# TNC HIGH CTL SPLIT
fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster2,
               data = go_rev[ go_rev$cluster=="TNC High", ])
survp <-  ggsurvplot(fit = fit, data = go_rev[ go_rev$cluster=="TNC High", ], 
                     risk.table = T, 
                     risk.table.height = 0.3,
                     legend.labs = c("CTL High", "CTL Low"),
                     palette = c("darkorange1", "seagreen4"),
                     pval = TRUE, 
                     pval.coord = c(6000, 0.9),
                     ggtheme = theme_survminer(),
                     #xlim=c(0,5000), #
                     #break.x.by = 2000,
                     title = c("TNC High & CTL Clustering"),
                     font.main = c(12, "bold", "darkblue"),
                     font.x = c(12, "bold.italic", "darkred"),
                     font.y = c(12, "bold.italic", "darkred"),
                     font.tickslab = c(12, "plain", "darkgreen"),
                     xlab = c("Time (days)"),
                     tables.y.text =F,
                     
                     facet.by = "Project.ID",
                     ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))

ggsave(filename =  paste("../output/", "NU_Reverse_Split_TNC_High.png", sep = ""),
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40,
       plot = print( survp))

# TNC LOW CTL SPLIT
fit <- survfit(Surv( times, vital_status ) ~ Project.ID + cluster2,
               data = go_rev[ go_rev$cluster=="TNC Low", ])
survp <-  ggsurvplot(fit = fit, data = go_rev[ go_rev$cluster=="TNC Low", ], 
                     legend.labs = c("CTL High", "CTL Low"),
                     palette = c("darkorange1", "seagreen4"),
                     risk.table = T, 
                     risk.table.height = 0.3,
                     pval = TRUE, 
                     pval.coord = c(6000, 0.9),
                     ggtheme = theme_survminer(),
                     #xlim=c(0,5000), #
                     #break.x.by = 2000,
                     title = c("TNC Low & TNC Clustering"),
                     font.main = c(12, "bold", "darkblue"),
                     font.x = c(12, "bold.italic", "darkred"),
                     font.y = c(12, "bold.italic", "darkred"),
                     font.tickslab = c(12, "plain", "darkgreen"),
                     xlab = c("Time (days)"),
                     tables.y.text =F,
                     
                     facet.by = "Project.ID",
                     ncol=2) %>% ggpar(font.main = c(12, "bold", "darkblue"),
                                       font.x = c(12, "bold.italic", "darkred"),
                                       font.y = c(12, "bold.italic", "darkred"),
                                       font.legend = c(12, "plain", "black"))

ggsave(filename =  paste("../output/", "NU_Reverse_Split_TNC_Low.png", sep = ""),
       device = "png", units = c("cm"), dpi = 300, width = 25, height = 40,
       plot = print( survp))

```