oncoprint2.R

library(readxl)


# read in spreadsheet
variants.filename <- dir(data_outside_app_dir, pattern = glob2rx("20*_pdx_dna_variants.xlsx"))
if (length(variants.filename) != 1) stop("too few or too many _pdx_dna_variants.xlsx files in dropbox")
# var_native <- read_excel(file.path(data_outside_app_dir,variants.filename),sheet=1)
variants <- read_excel(file.path(data_outside_app_dir, variants.filename), sheet = 1, col_types = "text")
variants <- as.data.frame(variants)
vs <- variants


# notes:
# tumor_sample_name_new is PDX_Name -- it is truncated and becomes pdx_id in database
# for True Mutation: 0 = benign, 1 = pathologic, 2 = vus

# produce desired form:
cols_keep <- c(
  "tumor_sample_name_new",
  "Canonical_Variant_Classification", # table 1
  "Canonical_Hugo_Symbol", # table 1
  "BestEffect_Variant_Classification", # table 2
  "BestEffect_Hugo_Symbol", # table 2
  "True_Mutation", # want to show just 1s or 1s+2s
  "allele_fraction",
  "Coverage"
)

v <- vs[cols_keep]
# v = v[v$True_Mutation != 0 & !is.na(v$True_Mutation),]
v <- v[v$True_Mutation == 1 & !is.na(v$True_Mutation), ]

v1 <- v[c(1, 4:8)]

#### --- recode *_Variant_Classification -- ####

# set up constants
snv <- c("Missense_Mutation", "Nonsense_Mutation", "Nonstop_Mutation", "Translation_Start_Site", "Missense")
indel <- c("Frame_Shift_Del", "Frame_Shift_Ins", "In_Frame_Del", "In_Frame_Ins")
splice <- c("Splice_Region", "Splice_Site")
mut_types_keep <- c(snv, indel, splice)

# - for v1 - #
# filter out terms that don't exist, like "0", "intron","intergenic_variant".
v1 <- v1[v1$BestEffect_Variant_Classification %in% mut_types_keep, ]
# Create new columns that map these classifications to the above terms: snv, indel, splice
v1$class2 <- NA_character_
for (i in 1:nrow(v1)) {
  class1 <- v1$BestEffect_Variant_Classification[i]
  if (class1 %in% snv) {
    v1$class2[i] <- "snv"
  } else if (class1 %in% indel) {
    v1$class2[i] <- "indel"
  } else if (class1 %in% splice) {
    v1$class2[i] <- "splice"
  } else {
    stop("Encountered unexpected value.")
  }
}
v1$class2 <- as.factor(v1$class2)
table(v1$BestEffect_Variant_Classification, v1$class2)

# # - for v2 - #
# # filter out terms that don't exist, like "0", "intron","intergenic_variant".
# v2 <- v2[v2$BestEffect_Variant_Classification %in% mut_types_keep,]
# # Create new columns that map these classifications to the above terms: snv, indel, splice
# v2$class2 <- NA_character_
# for(i in 1:nrow(v2)){
#   class1 <- v2$BestEffect_Variant_Classification[i]
#   if (class1 %in% snv){
#     v2$class2[i] <- "snv"
#   } else if (class1 %in% indel){
#     v2$class2[i] <- "indel"
#   } else if (class1 %in% splice){
#     v2$class2[i] <- "splice"
#   } else {
#     stop("Encountered unexpected value.")
#   }
# }
# v2$class2 <- as.factor(v2$class2)
# table(v2$BestEffect_Variant_Classification,v2$class2)



##### --- pivot into genes x samples --- ####

# - v1 - #

library(dplyr)
v1a <- v1 %>%
  dplyr::group_by(tumor_sample_name_new, BestEffect_Hugo_Symbol) %>%
  dplyr::summarise(id = paste(class2, collapse = ";"))

# note this looks like: (TODO: is snv;snv;snv a problem?)
# A tibble: 858 x 3
# Groups:   tumor_sample_name_new [?]
# tumor_sample_name_new BestEffect_Hugo_Symbol          id
# <chr>                 <chr>       <chr>
#   1         CBAB-10855-V2                 EPHA6         snv
# 2         CBAB-10855-V2                 GATA2         snv
# 3         CBAB-10855-V2                  MLL3 snv;snv;snv
# 4         CBAB-10855-V2                  PAX5         snv
v1a <- as.data.frame(lapply(v1a, as.factor))
v1a$id <- as.character(v1a$id)

library(reshape2)
v1mat <- acast(v1a, BestEffect_Hugo_Symbol ~ tumor_sample_name_new, value.var = "id")

# produce ComplexHeatmap

if (F) {
  col <- c(snv = "red", indel = "blue", splice = "yellow")
  png(filename = "tmp/test_oncoprint.png", width = 25, height = 20, units = "in", res = 250)
  library(ComplexHeatmap)
  ComplexHeatmap::oncoPrint(v1mat,
    get_type = function(x) strsplit(x, ";")[[1]],
    alter_fun = list(
      snv = function(x, y, w, h) grid.rect(x, y, w * 0.9, h * 0.9, gp = gpar(fill = col["snv"], col = NA)),
      indel = function(x, y, w, h) grid.rect(x, y, w * 0.9, h * 0.4, gp = gpar(fill = col["indel"], col = NA)),
      splice = function(x, y, w, h) grid.rect(x, y, w * 0.5, h * 0.5, gp = gpar(fill = col["splice"], col = NA))
    ), col = col,
    show_column_names = TRUE
  )
  dev.off()
}




# - v2 - #

## TODO


# - add to dataframe - #

# Template: WT1 Missense NM_024426 c.679A>G p.S227G of 0.481013 in 158 reads
# <gene> <type> <transcriptid> <nt change> <AA change> of <A.F.> in <read support> reads.
# only keep True Mutation == 1

# remove NAs
vdf <- vs[!is.na(vs$True_Mutation), ]
# filter for true mutations, desired columns
annot_cols <- c(
  "tumor_sample_name_new", "BestEffect_Hugo_Symbol", "BestEffect_Variant_Classification",
  "BestEffect_Refseq_mRNA_Id", "BestEffect_cDNA_Change", "BestEffect_Protein_Change", "allele_fraction", "Coverage"
)
vdf <- vdf[vdf$True_Mutation == 1, annot_cols]
# filter out intergenic
vdf <- vdf[vdf$BestEffect_Variant_Classification != "intergenic_variant", ]

# clean up data
vdf$BestEffect_Variant_Classification <- gsub("_", " ", vdf$BestEffect_Variant_Classification)
vdf$BestEffect_Variant_Classification <- gsub(" Mutation", "", vdf$BestEffect_Variant_Classification)
vdf$BestEffect_Variant_Classification <- gsub("Del", "Deletion", vdf$BestEffect_Variant_Classification)
vdf$BestEffect_Variant_Classification <- gsub("Ins", "Insertion", vdf$BestEffect_Variant_Classification)
vdf$BestEffect_Variant_Classification <- gsub("Frame Shift", "Frameshift", vdf$BestEffect_Variant_Classification)
vdf$BestEffect_Variant_Classification <- gsub("In Frame", "In-Frame", vdf$BestEffect_Variant_Classification)

# filter out nonsensical allele_fractions
vdf$allele_fraction <- as.numeric(vdf$allele_fraction)
vdf <- vdf[between(vdf$allele_fraction, 0, 1), ]

# filter out rows that are completely NA
vdf <- vdf[apply(vdf, 1, function(row) sum(is.na(row))) != ncol(vdf), ]

attach(vdf)
library(scales)
tmp <- paste(BestEffect_Hugo_Symbol, BestEffect_Variant_Classification, BestEffect_Refseq_mRNA_Id, BestEffect_cDNA_Change, BestEffect_Protein_Change, "in", scales::percent(allele_fraction), "of", Coverage, "reads")
detach(vdf)
vdf2 <- vdf
vdf2$PDX_Molecular_Details <- tmp

# replace " NA " with " __ "
vdf2$PDX_Molecular_Details <- gsub(" NA ", " __ ", vdf2$PDX_Molecular_Details)

vdf2$pdx_id <- substring(vdf2$tumor_sample_name_new, 1, 10)
vdf2 <- vdf2[vdf2$pdx_id %in% df$namenum, ]


# group vdf2 by pdx_id and concatenate the following, pipe-separating:
# 1. the BestEffect Hugo Symbol as PDX Molecular Alterations Positive, and
# 2. the PDX Molecular Details as PDX_Molecular Details
vdf_summ <- vdf2 %>%
  group_by(pdx_id) %>%
  summarise(
    "PDX Molecular Alterations Positive" = paste0(BestEffect_Hugo_Symbol, collapse = " | "),
    "PDX Molecular Details" = paste0(PDX_Molecular_Details, collapse = " | ")
  )

# note: this code supersedes oncoprint.R, so I'm removing the need for that now in compile_upload.R
# TODO later: git rm oncoprint.R, etc.

# complete the merge, moving columns and visibility indexes accordingly

# merge
df <- merge(df, vdf_summ, by.x = "namenum", by.y = "pdx_id", all.x = T)
obInvisRet_ind <- obInvisRet_ind + 1 # accounts for new 'namenum' at column 1 (was near end)

# move 'name' columns from #1,2 out of visible range
cols_to_move <- c("PDX RNA-Seq Name", "namenum")
df <- moveMe(df, cols_to_move, "after", names(df)[obInvisRet_ind])
obInvisRet_ind <- obInvisRet_ind - length(cols_to_move)

# move new mutation columns into visible range in intuitive context
cols_to_move <- c("PDX Molecular Alterations Positive", "PDX Molecular Details")
df <- moveMe(df, cols_to_move, "after", "PDX HemoSeq")
obInvisRet_ind <- obInvisRet_ind + length(cols_to_move)



# Appendix: template example used to build this document

if (F) {
  mat <- read.table(textConnection(
    ",s1,s2,s3
    g1,snv;indel,snv,indel
    g2,,snv;indel,snv
    g3,snv,,indel;snv"
  ), row.names = 1, header = TRUE, sep = ",", stringsAsFactors = FALSE)
  mat <- as.matrix(mat)
  mat

  library(ComplexHeatmap)
  col <- c(snv = "red", indel = "blue")
  ComplexHeatmap::oncoPrint(mat,
    get_type = function(x) strsplit(x, ";")[[1]],
    alter_fun = list(
      snv = function(x, y, w, h) grid.rect(x, y, w * 0.9, h * 0.9, gp = gpar(fill = col["snv"], col = NA)),
      indel = function(x, y, w, h) grid.rect(x, y, w * 0.9, h * 0.4, gp = gpar(fill = col["indel"], col = NA))
    ), col = col,
    show_column_names = TRUE
  )
}


# Ad hoc filter to say 'none' for known negative PDX Molecular Alterations

# pdx_hemo_none <- readxl::read_excel(file.path(data_outside_app_dir,"20180924_pdx_hemoseq_negative_mutations_or_vus.xlsx"))
# pdx_hemo_none <- pdx_hemo_none$`PDX Name`
# if( length(setdiff(pdx_hemo_none,df$`PDX Name`)) != 0 ){
#   stop("not all pdx hemoseq negative samples are in df. Probably a PRMS-mismatch.")
# }
# #
# for (pdx in pdx_hemo_none) {
#   if(!is.na(df[df$`PDX Name`==pdx,"PDX Molecular Alterations Positive"]) |
#     is.na(df[df$`PDX Name`==pdx,"PDX Molecular Details"])) {
#     print(paste(pdx,"has some Molecular Alteration info and shouldn't"))
#   }
# }

pdx_hemo_none <- df[df$`PDX HemoSeq` == "Complete" &
  !is.na(df$`PDX HemoSeq`) &
  is.na(df$`PDX Molecular Alterations Positive`) &
  is.na(df$`PDX Molecular Details`), ]$`PDX Name`

for (pdx in pdx_hemo_none) {
  df[df$`PDX Name` == pdx, "PDX Molecular Alterations Positive"] <- "none detected"
  df[df$`PDX Name` == pdx, "PDX Molecular Details"] <- "none detected"
}