README.md

metadata

license: mit
language:
  - en
tags:
  - genomics
  - transcription-factors
  - yeast
  - saccharomyces-cerevisiae
pretty_name: Yeast Dual Threshold Optimization Analysis
features:
  - applies_to:
      - dto
    fields:
      - name: pr_ranking_column
        dtype:
          class_label:
            - log2fc
            - pvalue
        description: >-
          Describes which column, effect (log2fc) or pvalue was used in ranking
          the perturbation response data. see scripts/dto_preparation.R for more
          info on each dataset
      - name: binding_rank_threshold
        dtype: float64
        description: >-
          Rank threshold used for the binding dataset in the DTO analysis. This
          represents the rank cutoff that maximizes overlap significance between
          binding and perturbation datasets.
        role: quantitative_measure
      - name: perturbation_rank_threshold
        dtype: float64
        description: >-
          Rank threshold used for the perturbation dataset in the DTO analysis.
          This represents the rank cutoff that maximizes overlap significance
          between binding and perturbation datasets.
        role: quantitative_measure
      - name: binding_set_size
        dtype: float64
        description: >-
          Number of targets in the binding dataset at the optimal rank
          threshold. This is the size of the set used to calculate overlap with
          the perturbation dataset.
        role: quantitative_measure
      - name: perturbation_set_size
        dtype: float64
        description: >-
          Number of targets in the perturbation dataset at the optimal rank
          threshold. This is the size of the set used to calculate overlap with
          the binding dataset.
        role: quantitative_measure
      - name: dto_fdr
        dtype: float64
        description: >-
          False discovery rate (FDR) for the dual threshold optimization test.
          Lower values indicate more significant overlap between binding and
          perturbation target sets. Missing values (NA) indicate insufficient
          data for DTO analysis.
        role: quantitative_measure
      - name: dto_empirical_pvalue
        dtype: float64
        description: >-
          Empirical p-value from permutation testing for the dual threshold
          optimization. This represents the probability of observing the
          observed overlap by chance. Missing values (NA) indicate insufficient
          data for DTO analysis.
        role: quantitative_measure
configs:
  - config_name: dto
    description: >-
      Dual threshold optimization analysis results comparing transcription
      factor binding experiments with gene expression perturbation experiments.
      The dataset is partitioned by binding dataset and perturbation dataset,
      allowing efficient queries for specific binding-perturbation comparisons.
    dataset_type: comparative
    data_files:
      - split: train
        path: dto/*/*/*.parquet
    dataset_info:
      features:
        - name: binding_id
          dtype: string
          description: >-
            Composite sample identifier for the binding experiment in format
            'repo_id;config_name;sample_id' (e.g.,
            'BrentLab/callingcards;annotated_feature_reprocess_intergenic_analysis;127')
          role: source_sample
        - name: perturbation_id
          dtype: string
          description: >-
            Composite sample identifier for the perturbation experiment in
            format 'repo_id;config_name;sample_id' (e.g.,
            'BrentLab/hackett_2020;hackett_2020_analysis_set;1504')
          role: source_sample
        - name: binding_repo_dataset
          dtype: string
          description: >-
            Partition key for the binding dataset: the repository name and
            config name joined by a hyphen (e.g., 'rossi_2021-macs_kang').
        - name: perturbation_repo_dataset
          dtype: string
          description: >-
            Partition key for the perturbation dataset: the repository name and
            config name joined by a hyphen (e.g.,
            'hackett_2020-hackett_2020_analysis_set').
      partitioning:
        enabled: true
        partition_by:
          - binding_repo_dataset
          - perturbation_repo_dataset
        path_template: >-
          dto/binding_repo_dataset={binding_repo_dataset}/perturbation_repo_dataset={perturbation_repo_dataset}/*.parquet

Yeast Comparative Analysis

Analyses that relate samples from different datasets in the BrentLab yeast collection. Each row relates two or more samples, which are referenced by composite identifiers of the form repo_id;config_name;sample_id, rather than describing the samples of a single dataset. Every config here has dataset_type: comparative.

The repository currently holds one analysis, dto, and can hold others as additional configs.

dto: dual threshold optimization

Dual threshold optimization (DTO) compares a binding dataset with a perturbation response dataset for the same regulator. For each pair of a binding sample and a perturbation sample it finds the rank thresholds on the two target lists that maximize the significance of their overlap, and reports the empirical p-value and false discovery rate of that overlap (dto_empirical_pvalue, dto_fdr). pr_ranking_column records whether the perturbation targets were ranked by effect (log2fc) or pvalue. See scripts/ for how the inputs are prepared and the results parsed.

The data is partitioned by binding dataset and perturbation dataset, one file per pair:

dto/binding_repo_dataset=<repo>-<config>/perturbation_repo_dataset=<repo>-<config>/part-0.parquet

Accessing Data

The examples below require labretriever (pip install labretriever) and/or the HuggingFace Hub client (pip install huggingface_hub).

Accessing Data with labretriever

This repository is part of a collection configured as a unified database using labretriever.VirtualDB. Download the collection config and use it to query the data directly in Python, or with an AI assistant using the labretriever plugin.

The collection config's links block maps each composite identifier field to the primary datasets it references. labretriever uses it to expose a dto_expanded view that splits binding_id into the primary dataset (binding_id_source) and the sample id (binding_id_id), and does the same for perturbation_id.

from labretriever.virtual_db import VirtualDB
from labretriever.datacard import DataCard

# Metadata
card = DataCard("BrentLab/yeast_comparative_analysis")
print([c.config_name for c in card.configs])  # list available datasets

# path to the downloaded brentlab_yeast_collection.yaml
vdb = VirtualDB("/path/to/brentlab_yeast_collection.yaml")

print(vdb.get_dataset_description("dto"))
vdb.query("SELECT * FROM dto_expanded LIMIT 5")

Direct parquet access

Use DataCard.info() to inspect available files, then download and query with DuckDB. The dto data is partitioned, so download a single binding and perturbation pair, or the whole directory.

from huggingface_hub import snapshot_download
import duckdb

repo_path = snapshot_download(
    repo_id="BrentLab/yeast_comparative_analysis",
    repo_type="dataset",
    allow_patterns="dto/**",
)
conn = duckdb.connect()
# returns a pandas DataFrame with the first 5 rows; the partition keys are
# available as columns because of hive_partitioning
conn.execute(
    "SELECT * FROM read_parquet(?, hive_partitioning = true) LIMIT 5",
    [f"{repo_path}/dto/*/*/*.parquet"],
).df()

Accessing using R

Clone the repository and read the partitioned dataset with arrow:

# install.packages("arrow")
arrow::open_dataset("dto") |> dplyr::collect()