scripts/genomecov_to_annotated_features.R

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

#' Convert BED format data frame to GRanges
#'
#' Handles coordinate system conversion from 0-indexed half-open BED format
#' to 1-indexed closed GenomicRanges format
#'
#' @param bed_df Data frame with chr, start, end columns in BED format (0-indexed, half-open)
#' @param zero_indexed Logical, whether input is 0-indexed (default: TRUE)
#' @return GRanges object
bed_to_granges <- function(bed_df, zero_indexed = TRUE) {

    if (!all(c("chr", "start", "end") %in% names(bed_df))) {
        stop("bed_df must have columns: chr, start, end")
    }

    # Convert from 0-indexed half-open [start, end) to 1-indexed closed [start, end]
    if (zero_indexed) {
        gr_start <- bed_df$start + 1
        gr_end <- bed_df$end
    } else {
        gr_start <- bed_df$start
        gr_end <- bed_df$end
    }

    # Create GRanges object (strand-agnostic for calling cards)
    gr <- GenomicRanges::GRanges(
        seqnames = bed_df$chr,
        ranges = IRanges::IRanges(start = gr_start, end = gr_end),
        strand = "*"
    )

    # Add any additional metadata columns
    extra_cols <- setdiff(names(bed_df), c("chr", "start", "end", "strand"))
    if (length(extra_cols) > 0) {
        GenomicRanges::mcols(gr) <- bed_df[, extra_cols, drop = FALSE]
    }

    return(gr)
}

#' Convert point-wise coverage to BED format
#'
#' @param coverage_df Data frame with chr, pos, pileup columns
#' @return Data frame in BED format with chr, start, end, score
coverage_to_bed <- function(coverage_df) {
    coverage_df %>%
        dplyr::rename(start = pos, score = pileup) %>%
        dplyr::mutate(end = start + 1) %>%  # pos is 0-indexed, end is exclusive
        dplyr::select(chr, start, end, score)
}

#' Sum scores of overlapping insertions per region
#'
#' @param insertions_gr GRanges object with insertions containing a 'score' metadata column
#' @param regions_gr GRanges object with regions
#' @return Numeric vector of summed scores per region
sum_overlap_scores <- function(insertions_gr, regions_gr) {
    # Find overlaps between regions and insertions
    overlaps <- GenomicRanges::findOverlaps(regions_gr, insertions_gr)

    # If no overlaps, return zeros
    if (length(overlaps) == 0) {
        return(rep(0, length(regions_gr)))
    }

    # Extract the scores for overlapping insertions
    scores <- GenomicRanges::mcols(insertions_gr)$score[S4Vectors::subjectHits(overlaps)]

    # Sum scores per region using tapply
    summed_scores <- tapply(scores, S4Vectors::queryHits(overlaps), sum)

    # Create result vector with zeros for regions without overlaps
    result <- rep(0, length(regions_gr))
    result[as.integer(names(summed_scores))] <- summed_scores

    return(result)
}

#' Combine replicates for a given regulator
#'
#' @param sample_set_id sample_id that identifies a set of replicates
#' @param genomecov_data List containing meta and ds (tagged experiment data)
#' @param regions_gr GRanges object with regions to quantify
combine_replicates_af <- function(sample_set_id, genomecov_data, regions_gr) {

    message(sprintf("Working on sample_id: %s", sample_set_id))

    run_accession_list <- genomecov_data$meta %>%
        filter(sample_id == sample_set_id) %>%
        pull(run_accession)

    library_totals <- genomecov_data$ds %>%
        filter(accession %in% run_accession_list) %>%
        group_by(accession) %>%
        summarise(n = sum(pileup, na.rm = TRUE)) %>%
        collect()

    replicate_region_counts <- map(run_accession_list, ~{
        run_acc <- .x

        coverage_gr <- genomecov_data$ds %>%
            filter(accession == run_acc) %>%
            collect() %>%
            coverage_to_bed() %>%
            bed_to_granges()

        sum_overlap_scores(coverage_gr, regions_gr)
    })

    replicates <- map2(replicate_region_counts, run_accession_list, ~{
        replicate_regions <- regions_gr
        replicate_regions$score <- .x
        replicate_regions
    })
    names(replicates) <- run_accession_list

    combined <- regions_gr
    combined$score <- Reduce(`+`, replicate_region_counts)

    list(
        library_total = library_totals,
        replicates = replicates,
        combined = combined
    )
}

#' Combine control samples
#'
#' @param genomecov_control List containing meta and ds (control data)
#' @param regions_gr GRanges object with regions to quantify
combine_control_af <- function(genomecov_control, regions_gr) {

    message("Processing control samples...")

    library_totals <- genomecov_control$ds %>%
        group_by(accession) %>%
        summarise(n = sum(pileup, na.rm = TRUE)) %>%
        collect()

    replicate_region_counts <- map(genomecov_control$meta$accession, ~{
        run_acc <- .x

        coverage_gr <- genomecov_control$ds %>%
            filter(accession == run_acc) %>%
            collect() %>%
            coverage_to_bed() %>%
            bed_to_granges()

        sum_overlap_scores(coverage_gr, regions_gr)
    })

    out <- regions_gr
    out$score <- Reduce(`+`, replicate_region_counts)

    list(
        library_totals = library_totals,
        af = out
    )
}

#' Calculate enrichment (calling cards effect)
#'
#' @param total_background_counts Total number of counts in background (scalar or vector)
#' @param total_experiment_counts Total number of counts in experiment (scalar or vector)
#' @param background_counts Number of counts in background per region (vector)
#' @param experiment_counts Number of counts in experiment per region (vector)
#' @param pseudocount Pseudocount to avoid division by zero (default: 0.1)
#' @return Enrichment values
calculate_enrichment <- function(total_background_counts,
                                 total_experiment_counts,
                                 background_counts,
                                 experiment_counts,
                                 pseudocount = 0.1) {

    # Input validation
    if (!all(is.numeric(c(total_background_counts, total_experiment_counts,
                          background_counts, experiment_counts)))) {
        stop("All inputs must be numeric")
    }

    # Get the length of the region vectors
    n_regions <- length(background_counts)

    # Ensure experiment_counts is same length as background_counts
    if (length(experiment_counts) != n_regions) {
        stop("background_counts and experiment_counts must be the same length")
    }

    # Recycle scalar totals to match region length if needed
    if (length(total_background_counts) == 1) {
        total_background_counts <- rep(total_background_counts, n_regions)
    }
    if (length(total_experiment_counts) == 1) {
        total_experiment_counts <- rep(total_experiment_counts, n_regions)
    }

    # Now check all are same length
    if (length(total_background_counts) != n_regions ||
        length(total_experiment_counts) != n_regions) {
        stop("All input vectors must be the same length or scalars")
    }

    # Calculate enrichment
    numerator <- experiment_counts / total_experiment_counts
    denominator <- (background_counts + pseudocount) / total_background_counts
    enrichment <- numerator / denominator

    # Check for invalid values
    if (any(enrichment < 0, na.rm = TRUE)) {
        stop("Enrichment values must be non-negative")
    }
    if (any(is.na(enrichment))) {
        stop("Enrichment values must not be NA")
    }
    if (any(is.infinite(enrichment))) {
        stop("Enrichment values must not be infinite")
    }

    return(enrichment)
}

#' Calculate Poisson p-values
#'
#' @param total_background_counts Total number of counts in background (scalar or vector)
#' @param total_experiment_counts Total number of counts in experiment (scalar or vector)
#' @param background_counts Number of counts in background per region (vector)
#' @param experiment_counts Number of counts in experiment per region (vector)
#' @param pseudocount Pseudocount for lambda calculation (default: 0.1)
#' @param ... additional arguments to `ppois`. note that lower tail is set to FALSE already
#' @return Poisson p-values
calculate_poisson_pval <- function(total_background_counts,
                                   total_experiment_counts,
                                   background_counts,
                                   experiment_counts,
                                   pseudocount = 0.1,
                                   ...) {

    # Input validation
    if (!all(is.numeric(c(total_background_counts, total_experiment_counts,
                          background_counts, experiment_counts)))) {
        stop("All inputs must be numeric")
    }

    # Get the length of the region vectors
    n_regions <- length(background_counts)

    # Ensure experiment_counts is same length as background_counts
    if (length(experiment_counts) != n_regions) {
        stop("background_counts and experiment_counts must be the same length")
    }

    # Recycle scalar totals to match region length if needed
    if (length(total_background_counts) == 1) {
        total_background_counts <- rep(total_background_counts, n_regions)
    }
    if (length(total_experiment_counts) == 1) {
        total_experiment_counts <- rep(total_experiment_counts, n_regions)
    }

    # Now check all are same length
    if (length(total_background_counts) != n_regions ||
        length(total_experiment_counts) != n_regions) {
        stop("All input vectors must be the same length or scalars")
    }

    # Calculate hop ratio
    hop_ratio <- total_experiment_counts / total_background_counts

    # Calculate expected number of counts (mu/lambda parameter)
    # Add pseudocount to avoid mu = 0
    mu <- (background_counts + pseudocount) * hop_ratio

    # Observed counts in experiment
    x <- experiment_counts

    # Calculate p-value: P(X >= x) = 1 - P(X < x) = 1 - P(X <= x-1)
    pval <- ppois(x - 1, lambda = mu, lower.tail = FALSE, ...)

    return(pval)
}

#' Calculate hypergeometric p-values
#'
#' @param total_background_counts Total number of counts in background (scalar or vector)
#' @param total_experiment_counts Total number of counts in experiment (scalar or vector)
#' @param background_counts Number of counts in background per region (vector)
#' @param experiment_counts Number of counts in experiment per region (vector)
#' @param ... additional arguments to phyper. Note that lower tail is set to false already
#' @return Hypergeometric p-values
calculate_hypergeom_pval <- function(total_background_counts,
                                     total_experiment_counts,
                                     background_counts,
                                     experiment_counts,
                                     ...) {

    # Input validation
    if (!all(is.numeric(c(total_background_counts, total_experiment_counts,
                          background_counts, experiment_counts)))) {
        stop("All inputs must be numeric")
    }

    # Get the length of the region vectors
    n_regions <- length(background_counts)

    # Ensure experiment_counts is same length as background_counts
    if (length(experiment_counts) != n_regions) {
        stop("background_counts and experiment_counts must be the same length")
    }

    # Recycle scalar totals to match region length if needed
    if (length(total_background_counts) == 1) {
        total_background_counts <- rep(total_background_counts, n_regions)
    }
    if (length(total_experiment_counts) == 1) {
        total_experiment_counts <- rep(total_experiment_counts, n_regions)
    }

    # Now check all are same length
    if (length(total_background_counts) != n_regions ||
        length(total_experiment_counts) != n_regions) {
        stop("All input vectors must be the same length or scalars")
    }

    # Hypergeometric parameters
    M <- total_background_counts + total_experiment_counts
    n <- total_experiment_counts
    N <- background_counts + experiment_counts
    x <- experiment_counts - 1

    # Handle edge cases
    valid <- (M >= 1) & (N >= 1)
    pval <- rep(1, length(M))

    # Calculate p-value for valid cases
    if (any(valid)) {
        pval[valid] <- phyper(x[valid], n[valid], M[valid] - n[valid], N[valid],
                              lower.tail = FALSE, ...)
    }

    return(pval)
}

#' Call peaks/quantify regions using calling cards approach
#'
#' @param sample_set_id sample_id that identifies a set of replicates
#' @param background_counts Vector of background counts per region
#' @param total_background_counts Total background counts (scalar)
#' @param annotated_feature_counts List of combined replicate data
#' @param regions_gr GRanges object with regions
#' @param pseudocount Pseudocount for calculations (default: 0.1)
#' @return List with replicates and combined quantifications
enrichment_analysis <- function(sample_set_id,
                                background_counts,
                                total_background_counts,
                                annotated_feature_counts,
                                regions_gr,
                                pseudocount = 0.1) {

    message(sprintf("Working on sample_id for %s", sample_set_id))

    counts_regulator <- annotated_feature_counts[[as.character(sample_set_id)]]

    replicate_quants <- map(names(counts_regulator$replicates), ~{
        message(sprintf("Working on replicate: %s", .x))
        gr <- counts_regulator$replicates[[.x]]

        af <- regions_gr

        experiment_counts <- gr$score
        total_experiment_counts <- counts_regulator$library_total %>%
            filter(accession == .x) %>%
            pull(n)

        # Add count columns
        GenomicRanges::mcols(af)$background_counts <- background_counts
        GenomicRanges::mcols(af)$experiment_counts <- experiment_counts
        GenomicRanges::mcols(af)$total_background_counts <- total_background_counts
        GenomicRanges::mcols(af)$total_experiment_counts <- total_experiment_counts

        # Calculate statistics
        GenomicRanges::mcols(af)$enrichment <- calculate_enrichment(
            total_background_counts = total_background_counts,
            total_experiment_counts = total_experiment_counts,
            background_counts = background_counts,
            experiment_counts = experiment_counts,
            pseudocount = pseudocount
        )

        GenomicRanges::mcols(af)$poisson_pval <- calculate_poisson_pval(
            total_background_counts = total_background_counts,
            total_experiment_counts = total_experiment_counts,
            background_counts = background_counts,
            experiment_counts = experiment_counts,
            pseudocount = pseudocount
        )

        GenomicRanges::mcols(af)$log_poisson_pval <- calculate_poisson_pval(
            total_background_counts = total_background_counts,
            total_experiment_counts = total_experiment_counts,
            background_counts = background_counts,
            experiment_counts = experiment_counts,
            pseudocount = pseudocount,
            log.p = TRUE
        )

        GenomicRanges::mcols(af)$hypergeometric_pval <- calculate_hypergeom_pval(
            total_background_counts = total_background_counts,
            total_experiment_counts = total_experiment_counts,
            background_counts = background_counts,
            experiment_counts = experiment_counts
        )

        GenomicRanges::mcols(af)$log_hypergeometric_pval <- calculate_hypergeom_pval(
            total_background_counts = total_background_counts,
            total_experiment_counts = total_experiment_counts,
            background_counts = background_counts,
            experiment_counts = experiment_counts,
            log.p = TRUE
        )

        # Calculate adjusted p-values
        GenomicRanges::mcols(af)$poisson_qval <- p.adjust(
            GenomicRanges::mcols(af)$poisson_pval, method = "fdr")
        GenomicRanges::mcols(af)$hypergeometric_qval <- p.adjust(
            GenomicRanges::mcols(af)$hypergeometric_pval, method = "fdr")

        af
    })

    names(replicate_quants) <- names(counts_regulator$replicates)

    message(sprintf("Working on the combined for sample_id %s", sample_set_id))

    combined_gr <- regions_gr

    combined_experiment_counts <- counts_regulator$combined$score
    combined_total_experiment_counts <- sum(counts_regulator$library_total$n)

    # Add count columns
    GenomicRanges::mcols(combined_gr)$background_counts <- background_counts
    GenomicRanges::mcols(combined_gr)$experiment_counts <- combined_experiment_counts
    GenomicRanges::mcols(combined_gr)$total_background_counts <- total_background_counts
    GenomicRanges::mcols(combined_gr)$total_experiment_counts <- combined_total_experiment_counts

    # Calculate statistics
    GenomicRanges::mcols(combined_gr)$enrichment <- calculate_enrichment(
        total_background_counts = total_background_counts,
        total_experiment_counts = combined_total_experiment_counts,
        background_counts = background_counts,
        experiment_counts = combined_experiment_counts,
        pseudocount = pseudocount
    )

    message("Calculating Poisson p-values...")
    GenomicRanges::mcols(combined_gr)$poisson_pval <- calculate_poisson_pval(
        total_background_counts = total_background_counts,
        total_experiment_counts = combined_total_experiment_counts,
        background_counts = background_counts,
        experiment_counts = combined_experiment_counts,
        pseudocount = pseudocount
    )

    GenomicRanges::mcols(combined_gr)$log_poisson_pval <- calculate_poisson_pval(
        total_background_counts = total_background_counts,
        total_experiment_counts = combined_total_experiment_counts,
        background_counts = background_counts,
        experiment_counts = combined_experiment_counts,
        pseudocount = pseudocount,
        log.p = TRUE
    )

    message("Calculating hypergeometric p-values...")
    GenomicRanges::mcols(combined_gr)$hypergeometric_pval <- calculate_hypergeom_pval(
        total_background_counts = total_background_counts,
        total_experiment_counts = combined_total_experiment_counts,
        background_counts = background_counts,
        experiment_counts = combined_experiment_counts
    )

    GenomicRanges::mcols(combined_gr)$log_hypergeometric_pval <- calculate_hypergeom_pval(
        total_background_counts = total_background_counts,
        total_experiment_counts = combined_total_experiment_counts,
        background_counts = background_counts,
        experiment_counts = combined_experiment_counts,
        log.p = TRUE
    )

    # Calculate adjusted p-values
    message("Calculating adjusted p-values...")
    GenomicRanges::mcols(combined_gr)$poisson_qval <- p.adjust(
        GenomicRanges::mcols(combined_gr)$poisson_pval, method = "fdr")
    GenomicRanges::mcols(combined_gr)$hypergeometric_qval <- p.adjust(
        GenomicRanges::mcols(combined_gr)$hypergeometric_pval, method = "fdr")

    message("Analysis complete!")

    list(
        replicates = replicate_quants,
        combined = combined_gr
    )
}

# ============================================================================
# Main analysis workflow
# ============================================================================

# Load data
genomic_features <- arrow::read_parquet(
    "~/code/hf/yeast_genome_resources/brentlab_features.parquet")

genomecov <- list(
    tagged = list(
        meta = arrow::read_parquet("~/code/hf/rossi_2021/rossi_2021_metadata.parquet"),
        ds = arrow::open_dataset("~/code/hf/rossi_2021/genome_map")
    ),
    control = list(
        meta = arrow::read_parquet("~/code/hf/rossi_2021/genome_map_control_meta.parquet"),
        ds = arrow::open_dataset("~/code/hf/rossi_2021/genome_map_control")
    )
)

# Get unique regulators
sample_id_list <- genomecov$tagged$meta %>%
    pull(sample_id) %>%
    unique()

# Load regions
regions_gr <- read_tsv(
    "~/code/hf/yeast_genome_resources/yiming_promoters.bed",
    col_names = c('chr', 'start', 'end', 'locus_tag', 'score', 'strand')) %>%
    bed_to_granges()

# Process control samples
rossi_2021_control <- combine_control_af(genomecov$control, regions_gr)

# Process all sample_id sets
annotated_feature_counts <- map(sample_id_list, ~{
    combine_replicates_af(.x, genomecov$tagged, regions_gr)
})
names(annotated_feature_counts) <- sample_id_list

# Perform enrichment analysis
annotated_feature_quants <- map(sample_id_list, ~{
    enrichment_analysis(
        .x,
        rossi_2021_control$af$score,
        sum(rossi_2021_control$library_totals$n),
        annotated_feature_counts,
        regions_gr
    )
})
names(annotated_feature_quants) <- sample_id_list

# Extract and format replicate-level results
annotated_features_quants_replicates <- map(annotated_feature_quants, ~{
    map(.x$replicates, as_tibble) %>%
        list_rbind(names_to = "run_accession")}) %>%
    list_rbind(names_to = "sample_id") %>%
    mutate(sample_id = as.integer(sample_id)) %>%
    left_join(
        genomecov$tagged$meta %>%
            ungroup() %>%
            select(sample_id, regulator_locus_tag, regulator_symbol, run_accession) %>%
            distinct(),
        by = c("sample_id", "run_accession")) %>%
    left_join(select(genomic_features, locus_tag, symbol)) %>%
    dplyr::rename(target_locus_tag = locus_tag, target_symbol = symbol) %>%
    dplyr::relocate(sample_id, run_accession, regulator_locus_tag, regulator_symbol,
                    target_locus_tag, target_symbol) %>%
    select(-c(score, width, strand))

# Write replicate-level results
# annotated_features_quants_replicates %>%
#     write_parquet(
#         "~/code/hf/rossi_2021/rossi_2021_af_replicates.parquet",
#         compression = "zstd",
#         write_statistics = TRUE,
#         chunk_size = 6708,
#         use_dictionary = c(
#             sample_id = TRUE,
#             run_accession = TRUE,
#             regulator_locus_tag = TRUE,
#             regulator_symbol = TRUE,
#             seqnames = TRUE,
#             target_locus_tag = TRUE,
#             target_symbol = TRUE
#         )
#     )

# Extract and format combined results
annotated_feature_quants_combined <- map(annotated_feature_quants, ~{
    as_tibble(.x$combined)}) %>%
    list_rbind(names_to = "sample_id") %>%
    mutate(sample_id = as.integer(sample_id)) %>%
    left_join(
        genomecov$tagged$meta %>%
            ungroup() %>%
            select(sample_id, regulator_locus_tag, regulator_symbol) %>%
            distinct(),
        by = "sample_id") %>%
    left_join(select(genomic_features, locus_tag, symbol)) %>%
    dplyr::rename(target_locus_tag = locus_tag, target_symbol = symbol) %>%
    dplyr::relocate(sample_id, regulator_locus_tag, regulator_symbol,
                    target_locus_tag, target_symbol) %>%
    select(-c(score, width, strand))

# Write combined results
# annotated_feature_quants_combined %>%
#     write_parquet(
#         "~/code/hf/rossi_2021/rossi_2021_af_combined.parquet",
#         compression = "zstd",
#         write_statistics = TRUE,
#         chunk_size = 6708,
#         use_dictionary = c(
#             sample_id = TRUE,
#             regulator_locus_tag = TRUE,
#             regulator_symbol = TRUE,
#             seqnames = TRUE,
#             target_locus_tag = TRUE,
#             target_symbol = TRUE
#         )
#     )