# Version info: R 4.2.2, Biobase 2.58.0, GEOquery 2.66.0, limma 3.54.0
################################################################
#   Data plots for selected GEO samples
library(GEOquery)
library(limma)
library(hgu219.db)
library(umap)
library(here)
library(tidyverse)

# load series and platform data from GEO

dir.create("data/gse65682")
gset <- getGEO("GSE65682", destdir = "data/gse65682", GSEMatrix = TRUE, getGPL = FALSE)
if (length(gset) > 1) idx <- grep("GPL13667", attr(gset, "names")) else idx <- 1
gset <- gset[[idx]]

ex <- exprs(gset)
# log2 transform
qx <- as.numeric(quantile(ex, c(0., 0.25, 0.5, 0.75, 0.99, 1.0), na.rm = T))
LogC <- (qx[5] > 100) ||
    (qx[6] - qx[1] > 50 && qx[2] > 0)
if (LogC) {
    ex[which(ex <= 0)] <- NaN
    ex <- log2(ex)
}

# box-and-whisker plot
dev.new(width = 3 + ncol(gset) / 6, height = 5)
par(mar = c(7, 4, 2, 1))
title <- paste("GSE65682", "/", annotation(gset), sep = "")
boxplot(ex, boxwex = 0.7, notch = T, main = title, outline = FALSE, las = 2)
dev.off()

# expression value distribution plot
par(mar = c(4, 4, 2, 1))
title <- paste("GSE65682", "/", annotation(gset), " value distribution", sep = "")
plotDensities(ex, main = title, legend = F)

# mean-variance trend
ex <- na.omit(ex) # eliminate rows with NAs
plotSA(lmFit(ex), main = "Mean variance trend, GSE65682")

# UMAP plot (multi-dimensional scaling)
ex <- ex[!duplicated(ex), ] # remove duplicates
ump <- umap(t(ex), n_neighbors = 15, random_state = 123)
plot(ump$layout, main = "UMAP plot, nbrs=15", xlab = "", ylab = "", pch = 20, cex = 1.5)

clean_char <- function(x, prefix_regex) {
    val <- str_squish(str_remove(x, prefix_regex))
    na_if(val, "NA")
}

pdat_mars <- as_tibble(pData(gset), rownames = "sample_id") |>
    select(sample_id, starts_with("characteristics"))

purrr::imap(pdat_mars |> select(-sample_id), ~ count(tibble(value = .x), value, name = "n") |> arrange(desc(n)))

gse65682_metadata <- as_tibble(pData(gset), rownames = "sample_id") |>
    mutate(
        healthy_control                = str_detect(title, "healthy"),
        title_date                     = str_extract(title, "\\d\\d_\\d{2}_\\d{4}"),
        title_identifier1              = str_extract(title, "(?<=\\d{2}_\\d{2}_\\d{4}_)\\w\\d+"),
        title_identifier2              = str_extract(title, "\\d+(?=\\])"),
        sex                            = clean_char(characteristics_ch1, "^gender:\\s*"),
        age                            = as.numeric(clean_char(characteristics_ch1.1, "^age:\\s*")),
        pneumonia_diagnosis            = clean_char(characteristics_ch1.2, "^pneumonia diagnoses:\\s*"),
        thrombocytopenia               = clean_char(characteristics_ch1.3, "^thrombocytopenia:\\s*"),
        endotype_cohort                = clean_char(characteristics_ch1.4, "^endotype_cohort:\\s*"),
        endotype_class                 = clean_char(characteristics_ch1.5, "^endotype_class:\\s*"),
        mortality_28d                  = as.logical(as.integer(clean_char(characteristics_ch1.6, "^mortality_event_28days:\\s*"))),
        time_to_event_28d              = as.numeric(clean_char(characteristics_ch1.7, "^time_to_event_28days:\\s*")),
        icu_acquired_infection         = clean_char(characteristics_ch1.8, "^icu_acquired_infection:\\s*"),
        icu_acquired_infection_paired  = clean_char(characteristics_ch1.9, "^icu_acquired_infection_paired:\\s*"),
        diabetes_mellitus              = clean_char(characteristics_ch1.10, "^diabetes_mellitus:\\s*"),
        abdominal_sepsis_or_control    = clean_char(characteristics_ch1.11, "^abdominal_sepsis_and_controls:\\s*")
    ) |>
    dplyr::select(
        sample_id, title_identifier1, title_identifier2, healthy_control, sex,
        age, pneumonia_diagnosis, thrombocytopenia, endotype_cohort, endotype_class,
        mortality_28d, time_to_event_28d, icu_acquired_infection, icu_acquired_infection_paired,
        diabetes_mellitus, abdominal_sepsis_or_control
    )

stopifnot(gse65682_metadata |> filter(!is.na(endotype_class)) |> count(is.na(mortality_28d)) |> pull(n) == 479)

your_probes_mars <- rownames(exprs(gset))
biocu219_probes <- keys(hgu219.db, keytype = "PROBEID")

length(your_probes_mars)
length(biocu219_probes)
length(intersect(your_probes_mars, biocu219_probes))
stopifnot(mean(!your_probes_mars %in% biocu219_probes) == 0)

anno_u219 <- AnnotationDbi::select(
    hgu219.db,
    keys = your_probes_mars,
    columns = c("SYMBOL", "ENTREZID", "ENSEMBL", "GENENAME", "UNIPROT"),
    keytype = "PROBEID"
)

gse65682_feature_metadata <- anno_u219 |>
    group_by(PROBEID) |>
    reframe(
        ENSEMBL = paste(unique(na.omit(ENSEMBL)), collapse = ";"),
        ENTREZID = dplyr::first(ENTREZID),
        SYMBOL = dplyr::first(SYMBOL),
        GENENAME = dplyr::first(GENENAME),
        UNIPROT = paste(unique(na.omit(UNIPROT)), collapse = ";")
    ) |>
    ungroup() |>
    mutate(across(c(ENSEMBL, UNIPROT), ~ na_if(.x, ""))) |>
    rename(feature_id = PROBEID) |>
    right_join(tibble(feature_id = your_probes_mars), by = "feature_id")

stopifnot(nrow(gse65682_feature_metadata) == length(your_probes_mars))
stopifnot(!any(duplicated(gse65682_feature_metadata$feature_id)))

mean(is.na(gse65682_feature_metadata$SYMBOL)) # for comparison against the other datasets' rates

gse65682_expr_long <- as_tibble(exprs(gset), rownames = "feature_id") |>
    pivot_longer(-feature_id, names_to = "sample_id", values_to = "value") |>
    arrange(sample_id, feature_id)

stopifnot(nrow(gse65682_expr_long) == nrow(exprs(gset)) * ncol(exprs(gset)))
stopifnot(all(unique(gse65682_expr_long$sample_id) %in% gse65682_metadata$sample_id))
stopifnot(all(unique(gse65682_expr_long$feature_id) %in% gse65682_feature_metadata$feature_id))

stopifnot(!any(duplicated(gse65682_metadata$sample_id)))
stopifnot(nrow(gse65682_metadata) == 802)

dir.create("data/gse65682/parquet", recursive = TRUE, showWarnings = FALSE)

arrow::write_parquet(gse65682_metadata, "data/gse65682/parquet/sample_metadata.parquet")
arrow::write_parquet(gse65682_feature_metadata, "data/gse65682/parquet/feature_metadata.parquet")
arrow::write_parquet(gse65682_expr_long, "data/gse65682/parquet/expression.parquet")