scripts/parse_harbison_data.R

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

# genomic feature harmonization table ----
# see https://huggingface.co/datasets/BrentLab/yeast_genome_resources
genes_table = arrow::open_dataset(here("data/genome_files/hf/features")) %>%
  as_tibble()

# function to read in the harbison raw data ----
read_in_harbison_data = function(xls_path, sheet_name, values_colname, n_max=6229){
  read_excel(xls_path,
             sheet=sheet_name,
             skip=1,
             n_max = n_max) %>%
    dplyr::rename(locus_tag=`...1`,
                  gene_name=`...2`,
                  description=`...3`) %>%
    mutate(across(!c(locus_tag, gene_name, description), as.numeric)) %>%
    pivot_longer(!c(locus_tag, gene_name, description),
                 names_to='tf_cond',
                 values_to=values_colname) %>%
    separate(tf_cond, c('tf', 'condition'), sep="_")
}

# read in the raw pvalue data ----
# note that n_max is set to 6229, which is the end of the genomic data. on some
# sheets, there is data accidently left in by the original researcher
harbison_pval_df_list = list(
  ypd = read_in_harbison_data(
    here('data/harbison/pvalbygene_forpaper_abbr.xls'),
    'YPD',
    'pvalue'),
  other_conds = read_in_harbison_data(
    here('data/harbison/pvalbygene_forpaper_abbr.xls'),
    "other conditions",
    "pvalue")
)

harbison_pval_df = bind_rows(harbison_pval_df_list)

# read in the raw effect ratio data ----
harbison_effect_df_list = list(
  ypd = read_in_harbison_data(
    here('data/harbison/ratiobygene_forpaper_abbr.xls'),
    'YPD',
    "effect"),
  other_conds = read_in_harbison_data(
    here('data/harbison/ratiobygene_forpaper_abbr.xls'),
    'Other Conditions',
    'effect'))

harbison_effect_df = bind_rows(harbison_effect_df_list)

# combine the effect and pvalue data ----
combined_harbison = harbison_effect_df %>%
  select(-gene_name) %>%
  left_join(select(harbison_pval_df, -gene_name)) %>%
  # remove the control sample and remove locus tags that correspond to
  # now deleted ORFs (these weren't merged into other annotations, they were
  # simply complely removed from the annotation set)
  filter(!locus_tag %in% c('YCL006C', 'YCR103C', 'YER187W-A')) %>%
  dplyr::rename(harbison_locus_tag = locus_tag) %>%
  mutate(tf = trimws(toupper(tf)))

# create a lookup map from the harbison targets to SGD 3-1 ----
locus_tags = combined_harbison %>%
  select(harbison_locus_tag) %>%
  distinct() %>%
  left_join(genes_table, by = c("harbison_locus_tag" = "locus_tag")) %>%
  mutate(locus_tag = harbison_locus_tag) %>%
  select(harbison_locus_tag, locus_tag, symbol)

locus_tags_aliases = read_csv('data/harbison/locus_tags_aliases.csv') %>%
  left_join(genes_table, by = c('curr_notation'='locus_tag')) %>%
  dplyr::rename(locus_tag = curr_notation) %>%
  select(harbison_locus_tag, locus_tag, symbol)

locus_tags_complete = locus_tags %>%
  filter(!is.na(symbol)) %>%
  bind_rows(locus_tags_aliases) %>%
  dplyr::rename(target_locus_tag = locus_tag,
                target_symbol = symbol)

stopifnot(nrow(locus_tags_complete) == nrow(locus_tags))

stopifnot(setequal(unique(combined_harbison$harbison_locus_tag),
                   locus_tags_complete$harbison_locus_tag))

# create a lookup map from the regulator names to SGD 3-1 ----
tf_symbol_to_locus_tag_df = combined_harbison %>%
  select(tf) %>%
  mutate(tf = trimws(toupper(tf))) %>%
  distinct() %>%
  left_join(genes_table, by = c('tf' = 'symbol')) %>%
  select(tf, locus_tag) %>%
  dplyr::rename(regulator_locus_tag = locus_tag) %>%
  mutate(regulator_symbol = tf) %>%
  select(tf, regulator_symbol, regulator_locus_tag)

tf_locus_tag_to_locus_tag_df = tf_symbol_to_locus_tag_df %>%
  filter(!complete.cases(.)) %>%
  select(tf) %>%
  left_join(genes_table, by = c('tf' = 'locus_tag')) %>%
  filter(!is.na(symbol)) %>%
  select(tf, symbol) %>%
  dplyr::rename(regulator_symbol = symbol) %>%
  mutate(regulator_locus_tag = tf)

tf_name_df_na = read_csv("data/harbison/tf_name_aliases.csv") %>%
  left_join(genes_table) %>%
  # tf_main is the current SGD 3-1 symbol
  select(tf, tf_main, locus_tag) %>%
  dplyr::rename(regulator_symbol = tf_main, regulator_locus_tag = locus_tag) %>%
  select(tf, regulator_symbol, regulator_locus_tag)

harbison_tf_map = bind_rows(
  tf_symbol_to_locus_tag_df[complete.cases(tf_symbol_to_locus_tag_df),],
  tf_locus_tag_to_locus_tag_df,
  tf_name_df_na
)

stopifnot(setequal(harbison_tf_map$tf, unique(combined_harbison$tf)))

# create the harmonized data ----
combined_harbison_harmonized = combined_harbison %>%
  left_join(harbison_tf_map) %>%
  left_join(locus_tags_complete) %>%
  dplyr::rename(harbison_regulator=tf) %>%
  select(-description)

harbison_db_meta <- read_csv("data/promotersetsig_db_20251127.csv",
                             col_types = cols(composite_binding = col_character()))

db_id_map = harbison_db_meta %>%
  filter(source_name == "harbison_chip") %>%
  select(id, regulator_locus_tag, condition) %>%
  distinct() %>%
  dplyr::rename(db_id = id) %>%
  bind_rows(
    tibble(
      regulator_locus_tag = c("YSC0017"),
      db_id = 0,
      condition = c('YPD')
      ))

combined_harbison_harmonized_for_parquet = combined_harbison_harmonized %>%
  select(regulator_locus_tag, regulator_symbol,
         condition, target_locus_tag, target_symbol,
         effect, pvalue) %>%
  left_join(db_id_map) %>%
  arrange(db_id) %>%
  group_by(db_id) %>%
  mutate(sample_id = cur_group_id()) %>%
  ungroup() %>%
  select(sample_id, db_id, regulator_locus_tag, regulator_symbol,
         condition,
         target_locus_tag, target_symbol,
         effect, pvalue)


# combined_harbison_harmonized_for_parquet %>%
#   write_parquet("~/code/hf/harbison_2004/harbison_2004.parquet",
#                 compression = "zstd",
#                 write_statistics = TRUE,
#                 chunk_size = 6226,
#                 use_dictionary = c(
#                   sample_id = TRUE,
#                   condition = TRUE,
#                   regulator_locus_tag = TRUE,
#                   regulator_symbol = TRUE,
#                   target_locus_tag = TRUE,
#                   target_symbol = TRUE
#                 )
#   )
#
#
# combined_harbison_harmonized %>%
#   select(harbison_locus_tag, target_locus_tag) %>%
#   distinct() %>%
#   write_csv("~/code/hf/harbison_2004/scripts/harbison_locus_tag_to_target_locus_tag.csv")
#
# combined_harbison_harmonized %>%
#   select(harbison_regulator, regulator_locus_tag) %>%
#   distinct() %>%
#   write_csv("~/code/hf/harbison_2004/scripts/harbison_regulator_to_regulator_locus_tag.csv")