## NOTE: The data is currently on /lts/mblab/downloaded_data/barkai_checseq
## and the parquet dataset is on the brentlab-strides aws at s3://yeast-binding-perturbation-data/barkai_checkseq

library(tidyverse)
library(here)
library(arrow)

sacCer3_genome = rtracklayer::import("~/ref/sacCer3/ucsc/sacCer3.fa.gz", format="fasta")

sacCer3_seqnames = unlist(map(str_split(names(sacCer3_genome), " "), ~.[[1]]))

sacCer3_genome_df = tibble(
  seqnames = rep(sacCer3_seqnames, Biostrings::width(sacCer3_genome))
) %>%
  group_by(seqnames) %>%
  mutate(start = row_number()-1,
         end = row_number()) %>%
  ungroup()

retrieve_series_paths = function(series_id){
  sra_meta_path = file.path("data/barkai_checseq", series_id, "SraRunTable.csv")
  stopifnot(file.exists(sra_meta_path))
  df = read_csv(sra_meta_path)

  data_files = list.files(here("data/barkai_checseq", series_id), "*.txt.gz", full.names = TRUE)

  stopifnot(nrow(df) == length(data_files))

  names(data_files) = str_extract(basename(data_files), "GSM\\d+")

  list(
    meta = sra_meta_path,
    files = data_files
  )
}


add_genomic_coordinate = function(checseqpath){

  bind_cols(sacCer3_genome_df,
            data.table::fread(checseqpath, sep = "\t", col.names='pileup'))

}

process_checseq_files = function(file){

  add_genomic_coordinate(file) %>%
    filter(pileup != 0)
}

series_list = map(set_names(c("GSE179430", "GSE209631", "GSE222268")), retrieve_series_paths)

dataset_basepath = here("data/barkai_checseq/hf/genome_map")

# Create output directory
dir.create(dataset_basepath, recursive = TRUE, showWarnings = FALSE)

for (series_id in names(series_list)) {

  message(glue::glue("Processing series {series_id}"))

  for (accession_id in names(series_list[[series_id]]$files)) {

    message(glue::glue("  Processing {accession_id}"))

    df <- process_checseq_files(
      series_list[[series_id]]$files[[accession_id]]
    ) %>%
      mutate(accession = accession_id, series = series_id)

    df %>%
      group_by(seqnames) %>%
    write_dataset(
      path = dataset_basepath,
      format = "parquet",
      partitioning = c("series", "accession"),
      existing_data_behavior = "overwrite",
      compression = "zstd",
      write_statistics = TRUE,
      use_dictionary = c(
        seqnames = TRUE
      )
    )

    gc()
  }
}

# the following code was used to parse an entire series to DF and then save
# to a parquet dataset. that was too large and I chose the dataset partitioning
# instead.

# split_manipulation <- function(manipulation_str) {
#   parts <- str_split(manipulation_str, "::")[[1]]
#
#   if (length(parts) != 2) {
#     stop("Unexpected format. Expected 'LOCUS::TAGGED_CONSTRUCT'")
#   }
#
#   tagged_locus <- parts[1]
#   rhs <- parts[2]
#
#   # default
#   dbd_donor_symbol_str <- "none"
#   ortholog <- "none"
#
#   # Check for paralog DBD
#   if (str_detect(rhs, "-[A-Za-z0-9]+DBD-Mnase$")) {
#     dbd_donor_symbol_str <- toupper(str_remove(str_split(rhs, "-", simplify = TRUE)[[2]], "DBD"))
#   } else if (str_detect(rhs, "^K\\.lactis .*?-Mnase$")) {
#     ortholog <- rhs
#   }
#
#   list(
#     mnase_tagged_symbol = tagged_locus,
#     dbd_donor_symbol = dbd_donor_symbol_str,
#     ortholog_donor = ortholog
#   )
# }
#
#
# split_deletion <- function(deletion_str) {
#   parts <- str_split(deletion_str, "::", simplify = TRUE)
#
#   list(
#     paralog_deletion_symbol = parts[1],
#     paralog_resistance_cassette = if (ncol(parts) >= 2) parts[2] else "none"
#   )
# }
#
# split_construct_to_tibble = function(split_list){
#   background = list(background=split_list[[1]])
#   manipulation_list = split_manipulation(split_list[[2]])
#   deletion_list = split_deletion(tryCatch(split_list[[3]], error = function(e) "none"))
#
#   bind_cols(map(list(background, manipulation_list, deletion_list), as_tibble))
#
# }
#
#
# split_constructs <- function(s) {
#   s <- str_trim(s)
#   if (s == "" || is.na(s)) return(character(0))
#   # split on spaces ONLY when the next token starts a new locus "XYZ::"
#   split_geno = str_split(s, "\\s+(?=[A-Za-z0-9_.()\\-]+::)")[[1]]
#
#   bind_cols(tibble(genotype = s), split_construct_to_tibble(split_geno))
#
#
# }
#
# gse178430_parsed_meta = bind_cols(
#   select(gse178430_meta, `GEO_Accession (exp)`, strainid, Instrument) %>%
#     dplyr::rename(accession = `GEO_Accession (exp)`,
#                   instrument = Instrument),
#   bind_rows(map(gse178430_meta$genotype, split_constructs))
# )