Datasets:

BrentLab
/

hughes_2006

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

# Constants ----
genomicfeatures_PATH <- here("data/genome_files/hf/features")
HUGHES_DATA_DIR <- here("data/hughes_2006")

# Load genomic features harmonization table ----
# See https://huggingface.co/datasets/BrentLab/yeast_genome_resources
genomicfeatures <- arrow::open_dataset(genomicfeatures_PATH) %>%
  as_tibble()

#' Read and process Hughes normalized data
#'
#' @param path Path to Excel file
#' @param ko Logical indicating if this is knockout data (TRUE) or overexpression (FALSE)
#' @return Tibble with processed data
read_hughes_normalized_data <- function(path, ko = TRUE) {
  if (!file.exists(path)) {
    stop("File not found: ", path)
  }
  df <- read_excel(path)
  # Standardize first column name
  colnames(df)[1] <- "hughes_target"
  # Pivot to long format
  df_long <- df %>%
    pivot_longer(
      cols = -hughes_target,
      names_to = "sample",
      values_to = "norm_log2fc"
    ) %>%
    mutate(norm_log2fc = as.numeric(norm_log2fc)) %>%
    # remove deleted ORFs
    filter(!hughes_target %in% c(
      "YCL006C",
      "YCR103C",
      "YER187W-A"
    )) %>%
    # these are transposons/retrotransposons and just have
    # typos
    mutate(hughes_target = case_when(
      hughes_target == "YDR034CD01" ~ "YDR034C-D",
      hughes_target == "YDR210WD01" ~ "YDR210W-D",
      hughes_target == "YDR261CD01" ~ "YDR261C-D",
      hughes_target == "YGR161CD01" ~ "YGR161C-D",
      hughes_target == "YPR158CD01" ~ "YPR158C-D",
      .default = hughes_target
    ))
  # Process based on data type
  if (ko) {
    df_long = df_long %>%
      separate(sample, into = c("prefix", "suffix"), sep = "/", remove = FALSE) %>%
      mutate(
        dye_orientation = str_extract(suffix, "[+-]$"),
        regulator_symbol = str_to_upper(str_remove(suffix, "[+-]"))
      ) %>%
      select(-prefix, -suffix)
  } else {
    df_long = df_long %>%
      mutate(
        dye_orientation = str_extract(sample, "[+-]$"),
        regulator_symbol = str_to_upper(str_remove(sample, "(-A)?OE[+-]"))
      )
  }

  df_long %>%
    select(-sample) %>%
    pivot_wider(id_cols = c(regulator_symbol, hughes_target),
                names_from = dye_orientation,
                values_from = norm_log2fc) %>%
    rename(dye_plus = `+`, dye_minus = `-`) %>%
    rowwise() %>%
    mutate(
      mean_norm_log2fc = case_when(
        # Both values are NA
        is.na(dye_plus) & is.na(dye_minus) ~ NA_real_,
        # Only one value is available - use that value
        is.na(dye_plus) & !is.na(dye_minus) ~ dye_minus,
        !is.na(dye_plus) & is.na(dye_minus) ~ dye_plus,
        # Both values available but have opposite signs - average to zero
        !is.na(dye_plus) & !is.na(dye_minus) & (sign(dye_plus) != sign(dye_minus)) ~ 0,
        # Both values available with same sign - take the mean
        TRUE ~ mean(c(dye_plus, dye_minus), na.rm = TRUE)
      )
    ) %>%
    ungroup()
}


# Load and process data ----
message("Loading knockout data...")
df_ko <- read_hughes_normalized_data(
  file.path(HUGHES_DATA_DIR, "Del_NormalizedRatios.xls"),
  ko = TRUE
)

message("Loading overexpression data...")
df_oe <- read_hughes_normalized_data(
  file.path(HUGHES_DATA_DIR, "OE_NormalizedRatios.xls"),
  ko = FALSE
)

message("Loading Z-scores...")
zscore_df <- read_excel(file.path(HUGHES_DATA_DIR, "Z_SCORES_FOR_106_EXPERIMENTS.xls")) %>%
  rename(hughes_target = `...1`) %>%
  pivot_longer(
    cols = -hughes_target,
    names_to = "sample",
    values_to = "zscore"
  ) %>%
  mutate(
    perturbation = case_when(
      str_detect(sample, "-D$") ~ "deletion",
      str_detect(sample, "^OE") ~ "overexpression",
      TRUE ~ "unknown"
    ),
    hughes_regulator = str_to_upper(str_remove(sample, "-D$|^OE"))
  )

# Load metadata ----
message("Loading metadata...")
hughes_2006_meta <- read_excel(file.path(HUGHES_DATA_DIR, "TRANSCRIPTION_FACTOR_LIST.xls")) %>%
  rename(
    regulator_locus_tag = Id_001,
    regulator_symbol = Id_002
  ) %>%
  mutate(
    essential = `Essential/Nonessential` == "essential",
    oe_passed_qc = `OE passed QC` == "yes",
    del_passed_qc = `DEL passed QC` == "yes"
  ) %>%
  select(-`Essential/Nonessential`, -ends_with("passed QC"))

# Data validation ----
message("Validating data consistency...")

validate_data_consistency <- function() {
  tests <- list(
    # Check regulator locus tags match genomic features
    regulator_locus_consistency = setequal(
      intersect(genomicfeatures$locus_tag, hughes_2006_meta$regulator_locus_tag),
      hughes_2006_meta$regulator_locus_tag
    ),

    # Check regulator symbols match genomic features
    regulator_symbol_consistency = setequal(
      intersect(genomicfeatures$symbol, hughes_2006_meta$regulator_symbol),
      hughes_2006_meta$regulator_symbol
    ),

    # Check OE regulators match metadata
    oe_regulator_consistency = setequal(
      intersect(hughes_2006_meta$regulator_symbol, unique(df_oe$regulator_symbol)),
      unique(df_oe$regulator_symbol)
    ),

    # Check KO regulators match metadata
    ko_regulator_consistency = setequal(
      intersect(hughes_2006_meta$regulator_symbol, unique(df_ko$regulator_symbol)),
      unique(df_ko$regulator_symbol)
    ),

    # Check target consistency between KO and OE
    target_consistency = setequal(
      unique(df_ko$hughes_target),
      unique(df_oe$hughes_target)
    ),

    # Check targets match genomic features
    target_genomic_consistency = setequal(
      unique(df_oe$hughes_target),
      intersect(unique(df_oe$hughes_target), genomicfeatures$locus_tag)
    )
  )

  # Report results
  failed_tests <- names(tests)[!unlist(tests)]

  if (length(failed_tests) == 0) {
    message("✓ All validation tests passed")
  } else {
    stop("✗ Validation failed for: ", paste(failed_tests, collapse = ", "))
  }

  invisible(tests)
}

validate_data_consistency()

# Summary statistics ----
message("Data loading complete. Summary:")
message("- Knockout experiments: ", length(unique(df_ko$regulator_symbol)), " regulators")
message("- Overexpression experiments: ", length(unique(df_oe$regulator_symbol)), " regulators")
message("- Target genes: ", length(unique(df_ko$hughes_target)))
message("- Z-score experiments: ", length(unique(zscore_df$sample)))

# Note about missing targets in Z-score data
missing_targets_count <- length(setdiff(unique(df_ko$hughes_target),
                                        unique(zscore_df$hughes_target)))
if (missing_targets_count > 0) {
  message("- Z-score data missing ", missing_targets_count, " targets present in KO/OE data")
}

missing_locus_tags <- setdiff(
  unique(df_oe$hughes_target),
  intersect(unique(df_oe$hughes_target), genomicfeatures$locus_tag)
)

genome_map = tibble(
  locus_tag = intersect(unique(df_oe$hughes_target),
                        genomicfeatures$locus_tag)) %>%
  left_join(genomicfeatures %>% select(locus_tag, symbol)) %>%
  mutate(hughes_target = locus_tag) %>%
  bind_rows(
    genomicfeatures %>%
      filter(str_detect(alias, paste(missing_locus_tags, collapse = "|"))) %>%
      mutate(alias_match = str_extract(alias, paste(missing_locus_tags, collapse = "|"))) %>%
      dplyr::rename(hughes_target = alias_match) %>%
      select(hughes_target, locus_tag, symbol)
  ) %>%
  dplyr::rename(target_locus_tag = locus_tag, target_symbol = symbol)

stopifnot(setequal(genome_map$hughes_target, unique(df_oe$hughes_target)))

hughes_2006_meta_with_id = hughes_2006_meta %>%
  arrange(regulator_locus_tag) %>%
  group_by(regulator_locus_tag) %>%
  mutate(sample_id = cur_group_id()) %>%
  ungroup() %>%
  select(sample_id, all_of(colnames(hughes_2006_meta))) %>%
  arrange(sample_id)

stopifnot(
  hughes_2006_meta_with_id %>% group_by(sample_id) %>% filter(n()>1) %>% nrow() == 0
)

df_oe_harmonized = df_oe %>%
  left_join(
    select(hughes_2006_meta_with_id,
           sample_id,
           regulator_locus_tag,
           regulator_symbol)) %>%
  left_join(genome_map) %>%
  select(sample_id,
         regulator_locus_tag, regulator_symbol,
         target_locus_tag, target_symbol,
         dye_plus, dye_minus, mean_norm_log2fc) %>%
  arrange(sample_id, target_locus_tag)

df_oe_harmonized %>%
  write_parquet("~/code/hf/hughes_2006/overexpression.parquet",
                compression = "zstd",
                write_statistics = TRUE,
                use_dictionary = c(
                  sample_id = TRUE,
                  regulator_locus_tag = TRUE,
                  regulator_symbol = TRUE,
                  target_locus_tag = TRUE,
                  target_symbol = TRUE
                )
  )

df_ko_harmonized = df_ko %>%
  left_join(
    select(hughes_2006_meta_with_id,
           sample_id,
           regulator_locus_tag,
           regulator_symbol)) %>%
  left_join(genome_map) %>%
  select(sample_id,
         regulator_locus_tag, regulator_symbol,
         target_locus_tag, target_symbol,
         dye_plus, dye_minus, mean_norm_log2fc) %>%
  arrange(sample_id, target_locus_tag)

df_ko_harmonized %>%
  write_parquet("~/code/hf/hughes_2006/knockout.parquet",
                compression = "zstd",
                write_statistics = TRUE,
                use_dictionary = c(
                  sample_id = TRUE,
                  regulator_locus_tag = TRUE,
                  regulator_symbol = TRUE,
                  target_locus_tag = TRUE,
                  target_symbol = TRUE
                )
  )

hughes_2006_meta_with_id %>%
  janitor::clean_names() %>%
  write_parquet("~/code/hf/hughes_2006/metadata.parquet",
                compression = "zstd",
                write_statistics = TRUE,
                use_dictionary = c(
                  sample_id = TRUE,
                  regulator_locus_tag = TRUE,
                  regulator_symbol = TRUE,
                  oe_passed_qc = TRUE,
                  del_passed_qc = TRUE
                )
  )