adding rossi original authors scores

README.md

@@ -75,11 +75,14 @@ features:
     dtype: string
     description: Standard gene symbol of the transcription factor
 
+
 - applies_to:
   - rossi_2021_af_replicates
   - rossi_2021_af_combined
   - rossi_2021_af_replicates_mindel
   - rossi_2021_af_combined_mindel
+  - yep_filtered_peaks_yiming
+  - yep_filtered_peaks_mindel
   fields:
   - name: target_locus_tag
     dtype: string
@@ -98,6 +101,13 @@ features:
   - name: end
     dtype: int64
     description: Promoter region end position (1-based, inclusive)
+
+- applies_to:
+  - rossi_2021_af_replicates
+  - rossi_2021_af_combined
+  - rossi_2021_af_replicates_mindel
+  - rossi_2021_af_combined_mindel
+  fields:
   - name: background_counts
     dtype: int64
     description: Read counts in the background/control sample for this peak region
@@ -146,6 +156,8 @@ features:
 - applies_to:
   - rossi_2021_af_replicates_mindel
   - rossi_2021_af_combined_mindel
+  - yep_filtered_peaks_yiming
+  - yep_filtered_peaks_mindel
   fields:
   - name: width
     dtype: int64
@@ -153,6 +165,11 @@ features:
   - name: strand
     dtype: string
     description: Genomic strand of the promoter region (+ or -)
+
+- applies_to:
+  - rossi_2021_af_replicates_mindel
+  - rossi_2021_af_combined_mindel
+  fields:
   - name: mindel_name
     dtype: string
     description: Name of the promoter region as defined in the Mindel promoters
@@ -169,6 +186,22 @@ features:
     description: >-
       this is a feature from the Mindel data that I have not documented. It is a TODO.
 
+- applies_to:
+  - yep_filtered_peaks_yiming
+  - yep_filtered_peaks_mindel
+  fields:
+  - name: score
+    dtype: float64
+    description: >-
+      The filtered bed file representing the high confidence ChExMix peaks from
+      yeastepigenome.org score. Where multiple peaks intersect the same promoter,
+      the median score is taken for that promoter. Replicates are then combined. When 
+      combining replicates, the median score across all replicates is taken. Note that
+      when a replicate does not have a peak score for a promoter in the promoter set,
+      that promoter is assigned a score of 0. So if replicate A as a peak score of
+      500 and replicate B did not have any peak intersecting with that promoter, then
+      the combined score for that promoter would be 250.
+
 configs:
 - config_name: rossi_2021_metadata_replicate
   description: Metadata describing the tagged regulator in each experiment
@@ -198,7 +231,7 @@ configs:
 - config_name: rossi_2021_metadata_sample
   description: Sample-level metadata for combined ChIP-exo experiments including experimental conditions
   dataset_type: metadata
-  applies_to: ["rossi_2021_af_combined", "rossi_2021_af_combined_mindel"]
+  applies_to: ["rossi_2021_af_combined", "rossi_2021_af_combined_mindel", "yep_filtered_peaks_yiming", "yep_filtered_peaks_mindel]
   data_files:
   - split: train
     path: rossi_2021_metadata_sample.parquet
@@ -343,7 +376,9 @@ configs:
       role: sample_id
 
 - config_name: rossi_2021_af_combined_mindel
-  description: Combined ChIP-exo annotated features with binding peaks and statistical significance metrics aggregated across biological replicates
+  description: >-
+    Combined ChIP-exo annotated features with binding peaks and statistical
+    significance metrics aggregated across biological replicates
   dataset_type: annotated_features
   genome_resources:
     region_sets:
@@ -359,6 +394,54 @@ configs:
       dtype: int64
       description: Unique identifier combining regulator and replicates
       role: sample_id
+
+- config_name: yep_filtered_peaks_yiming
+  description: >-
+    The filtered high confidence peaks reported at yeastepigenome.org intersected 
+    with the Kang promoter regions. see `scripts/parse_rossi_filtered_peaks.R`
+    for more details. NOTE: be sure to read the definition of the score column, 
+    which is at the top of this README in the common features section, as the
+    score has been combined across peaks intersecting the same promoter in the same
+    replicate, and then also across replicates.
+  dataset_type: annotated_features
+  genome_resources:
+    region_sets:
+      Kang:
+        path: https://huggingface.co/datasets/BrentLab/yeast_genome_resources/blob/main/yiming_promoters.bed
+        join_column: target_locus_tag
+  data_files:
+  - split: train
+    path: yep_filtered_peaks_yiming.parquet
+  dataset_info:
+    features:
+    - name: sample_id
+      dtype: int64
+      description: Unique identifier for the biological replicate
+      role: sample_id
+
+- config_name: yep_filtered_peaks_mindel
+  description: >-
+    The filtered high confidence peaks reported at yeastepigenome.org intersected 
+    with the Mindel promoter regions. see `scripts/parse_rossi_filtered_peaks.R`
+    for more details. NOTE: be sure to read the definition of the score column, 
+    which is at the top of this README in the common features section, as the
+    score has been combined across peaks intersecting the same promoter in the same
+    replicate, and then also across replicates.
+  dataset_type: annotated_features
+  genome_resources:
+    region_sets:
+      Mindel:
+        path: https://huggingface.co/datasets/BrentLab/yeast_genome_resources/blob/main/mindel_promoters.csv.gz
+        join_column: target_locus_tag
+  data_files:
+  - split: train
+    path: yep_filtered_peaks_mindel.parquet
+  dataset_info:
+    features:
+    - name: sample_id
+      dtype: int64
+      description: Unique identifier for the biological replicate
+      role: sample_id
 ---
 # Rossi 2021
 

scripts/parse_rossi_filtered_peaks.R

@@ -0,0 +1,212 @@
+# fetch the zip files from the yeastepigenome.org server
+# https://www.datacommons.psu.edu/download/eberly/pughlab/yeast-epigenome-project/
+# and then extract them
+# with
+#!/usr/bin/env bash
+# fetch_yep_peaks.sh
+#
+# Extracts *_chexmix_filtered_peaks.bed from each *_YEP.zip in ZIP_DIR.
+# Output: yep_filtered_peaks/ in the current working directory.
+#
+# set -euo pipefail
+#
+# ZIP_DIR="${1:-/lts/mblab/downloaded_data/chipexo/yeastepigenome_data}"
+# OUT_DIR="$(pwd)/yep_filtered_peaks"
+#
+# if [[ ! -d "$ZIP_DIR" ]]; then
+# echo "[error] ZIP_DIR not found: $ZIP_DIR" >&2
+# exit 1
+# fi
+#
+# mkdir -p "$OUT_DIR"
+#
+# COUNT=0; SKIP=0; FAIL=0
+#
+# for zip_path in "$ZIP_DIR"/*_YEP.zip; do
+# zip_name=$(basename "$zip_path")
+# sample_id="${zip_name%_YEP.zip}"
+# bed_name="${sample_id}_chexmix_filtered_peaks.bed"
+# dest="$OUT_DIR/$bed_name"
+#
+# if [[ -f "$dest" ]]; then
+# echo "[skip] $bed_name already exists"
+# (( SKIP++ )) || true
+# continue
+# fi
+#
+# # file path inside the zip may vary — find it
+# bed_in_zip=$(unzip -Z1 "$zip_path" 2>/dev/null | grep -m1 "${bed_name}$" || true)
+#
+# if [[ -z "$bed_in_zip" ]]; then
+# echo "[warn] $bed_name not found in $zip_name — skipping" >&2
+# (( FAIL++ )) || true
+# continue
+# fi
+#
+# echo "[ext ] $zip_name  →  $bed_name"
+# unzip -p "$zip_path" "$bed_in_zip" > "$dest"
+# (( COUNT++ )) || true
+# done
+#
+# echo ""
+# echo "[done] extracted=$COUNT  skipped=$SKIP  failed=$FAIL"
+# echo "[done] output in: $OUT_DIR"
+
+# the following can be run on the output
+
+library(tidyverse)
+library(arrow)
+library(GenomicRanges)
+library(here)
+
+#' Convert BED format data frame to GRanges
+#'
+#' Handles coordinate system conversion from 0-indexed half-open BED format
+#' to 1-indexed closed GenomicRanges format
+#'
+#' @param bed_df Data frame with chr, start, end columns in BED format (0-indexed, half-open)
+#' @param zero_indexed Logical, whether input is 0-indexed (default: TRUE)
+#' @return GRanges object
+bed_to_granges <- function(bed_df, zero_indexed = TRUE) {
+
+    if (!all(c("chr", "start", "end") %in% names(bed_df))) {
+        stop("bed_df must have columns: chr, start, end")
+    }
+
+    # Convert from 0-indexed half-open [start, end) to 1-indexed closed [start, end]
+    if (zero_indexed) {
+        gr_start <- bed_df$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)
+}
+
+#' Annotate genomic regions with ChExMix peak scores
+#'
+#' Intersects a set of genomic regions (e.g. promoters) with ChExMix filtered
+#' peaks and assigns each region a score. Regions overlapping a single peak
+#' receive that peak's score; regions overlapping multiple peaks receive the
+#' median score across all overlapping peaks. Non-overlapping regions are
+#' scored 0.
+#'
+#' @param peaks_df Data frame with BED-format peak calls (0-indexed, half-open)
+#'   and a \code{score} column, as produced by \code{read_in_filtered_bed}.
+#' @param regions_gr GRanges object of genomic regions to annotate
+#'   (e.g. an element of \code{regions_list}).
+#'
+#' @return A copy of \code{regions_gr} with an additional metadata column
+#'   \code{score}: the peak score for overlapping regions, 0 otherwise.
+#'
+#' @seealso \code{\link{bed_to_granges}}, \code{\link{read_in_filtered_bed}}
+peaks_to_promoter_scores <- function(peaks_df, regions_gr) {
+
+    peaks_gr <- bed_to_granges(peaks_df)
+
+    hits <- GenomicRanges::findOverlaps(regions_gr, peaks_gr)
+
+    # for each hit, pull the peak score
+    scores_df <- tibble(
+        region_idx = S4Vectors::queryHits(hits),
+        score      = peaks_df$score[S4Vectors::subjectHits(hits)]
+    ) |>
+        summarise(score = median(score), .by = region_idx)
+
+    # initialize all promoters to 0, fill in overlapping ones
+    result <- rep(0, length(regions_gr))
+    result[scores_df$region_idx] <- scores_df$score
+
+    regions_gr_tmp <- regions_gr
+    regions_gr_tmp$score <- result
+    regions_gr_tmp
+}
+
+
+rossi_meta = arrow::read_parquet("~/code/hf/rossi_2021/rossi_2021_metadata.parquet")
+chrmap = read_csv("~/code/hf/yeast_genome_resources/chrmap.csv.gz")
+
+brentlab_features = read_csv("~/code/hf/yeast_genome_resources/brentlab_features.csv.gz")
+
+# Load regions
+regions_list = list(
+    yiming = read_tsv(
+        "~/code/hf/yeast_genome_resources/yiming_promoters.bed",
+        col_names = c('chr', 'start', 'end', 'locus_tag', 'score', 'strand')) %>%
+        bed_to_granges(),
+    mindel = arrow::read_parquet("~/code/hf/yeast_genome_resources/mindel_promoters.parquet") |>
+        bed_to_granges()
+)
+
+filtered_bed = list.files(here("data/yep_filtered_peaks"), full.names = TRUE)
+names(filtered_bed) = str_extract(basename(filtered_bed), "^\\d+")
+
+read_in_filtered_bed = function(path){
+    read_tsv(path, col_names = c('chr', 'start', 'end', 'source', 'score')) |>
+        left_join(dplyr::select(chrmap, chr, ucsc)) |>
+        dplyr::select(-chr) |>
+        dplyr::rename(chr = ucsc) |>
+        dplyr::relocate(chr)
+}
+
+filtered_bed_df_list = map(filtered_bed, read_in_filtered_bed)
+
+filtered_bed_df_list_no_null = filtered_bed_df_list[unlist(map(filtered_bed_df_list, ~nrow(.) > 0))]
+
+filtered_bed_df_list_no_null_in_meta = filtered_bed_df_list_no_null[
+    as.integer(names(filtered_bed_df_list_no_null)) %in% rossi_meta$yeastepigenome_id]
+
+af_list = list(
+    yiming = map(filtered_bed_df_list_no_null_in_meta, peaks_to_promoter_scores, regions_list$yiming),
+    mindel = map(filtered_bed_df_list_no_null_in_meta, peaks_to_promoter_scores, regions_list$mindel)
+)
+
+af_list_dfs = list(
+    yiming = bind_rows(map(af_list$yiming, as_tibble), .id = "yeastepigenome_id") |>
+        mutate(yeastepigenome_id = as.integer(yeastepigenome_id)) |>
+        left_join(dplyr::select(brentlab_features, locus_tag, symbol)) |>
+        dplyr::rename(target_locus_tag = locus_tag, target_symbol = symbol) |>
+        left_join(rossi_meta) |>
+        dplyr::rename(seqnames = chr) |>
+        group_by(sample_id, seqnames, start, end, width, strand, target_locus_tag, target_symbol) |>
+        reframe(score = median(score)) |>
+        group_by(sample_id) |>
+        arrange(desc(score)),
+    mindel = bind_rows(map(af_list$mindel, as_tibble), .id = "yeastepigenome_id") |>
+        mutate(yeastepigenome_id = as.integer(yeastepigenome_id)) |>
+        left_join(rossi_meta) |>
+        dplyr::rename(seqnames = chr) |>
+        group_by(sample_id, seqnames, start, end, width, strand, target_locus_tag, target_symbol) |>
+        reframe(score = median(score)) |>
+        group_by(sample_id) |>
+        arrange(desc(score))
+)
+
+arrow::write_parquet(
+    af_list_dfs$yiming,
+    "~/code/hf/rossi_2021/yep_filtered_peaks_yiming.parquet",
+    compression = "zstd",
+    write_statistics = TRUE
+)
+
+arrow::write_parquet(
+    af_list_dfs$mindel,
+    "~/code/hf/rossi_2021/yep_filtered_peaks_mindel.parquet",
+    compression = "zstd",
+    write_statistics = TRUE
+)

yep_filtered_peaks_mindel.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f8473129d802f994ddc01c9e2d5d1625da8b541dc3439b7a389397568a93662
+size 24383507

yep_filtered_peaks_yiming.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbd1506072d97198bce01b497cb93a33a79772843bd6fa22bfe5dd04e3037795
+size 30745732