DANDELION is an R package for identifying disease-proximal genes (DPGs) that may mediate the effects of disease-associated loci on disease risk. It provides a causal mediation–inspired gene prioritization framework that integrates trans-regulatory association signals with gene-level disease association evidence.

This package accompanies the study:

Trans-regulatory gene mapping prioritizes disease drivers in asthma.

The release version of DANDELION is archived on Zenodo at DOI: 10.5281/zenodo.19911607.

In the R functions and examples below:

• gene1 denotes the disease-distal exposure, which can be either a regulatory gene or a SNP.
• gene2 denotes the disease-proximal candidate mediator and trans-regulatory target.

Overview of DANDELION R package

DANDELION combines two sources of evidence:

1. Trans-regulatory association p-values, representing the association between an upstream exposure and a downstream gene.
2. Gene-level disease association p-values, such as burden-test p-values from whole-exome sequencing (WES) or other gene-level association tests.

To assess mediation, DANDELION implements and adapts the Divide-Aggregate Composite-null Test (DACT) framework to the trans-gene regulation setting. Specifically, DANDELION combines trans-association p-values with gene-level burden-test p-values to prioritize candidate disease-proximal genes and distal-proximal gene pairs.

Details of the causal mediation assumptions, composite-null structure, p-value computation, and statistical theory are provided in the original DACT paper (Liu et al. 2022).

1. Gene-based DANDELION

In gene-based DANDELION, the exposure side contains disease-distal genes. The input trans-association matrix represents gene-to-gene trans-regulatory p-values.

A typical use case is:

disease-distal gene -> trans target gene -> disease trait

When raw genotype and gene expression data are available, gene-based trans-regulatory effects can be estimated using approaches such as GBAT. In this setting, predicted genetically regulated expression of a distal gene can be tested against genome-wide target genes to obtain gene-to-gene trans-association p-values.

2. SNP-based DANDELION

In SNP-based DANDELION, the exposure side contains SNPs. The input trans-association matrix represents SNP-to-gene trans-regulatory p-values, such as trans-eQTL summary statistics.

A typical use case is:

SNP -> trans target gene -> disease trait

This setting is useful when raw genotype and expression data are unavailable but trans-eQTL summary statistics are available.

Features

• Supports both gene-based and SNP-based DANDELION analyses.
• Integrates trans-regulatory p-values with gene-level disease association p-values.
• Removes local cis genes using a user-defined genomic window before trans-mediation testing.
• Applies a DACT-style composite-null testing framework to prioritize candidate mediators.
• Returns structured R objects that users can inspect, print, and extract from.
• Provides post-processing functions for organizing significant gene pairs.
• Supports locus-level grouping of nearby exposure genes.
• Provides a simple network visualization function for DANDELION-prioritized relationships.

Installation

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("mxxptian/DANDELION")

After installation, load the package with:

library(DANDELION)

Optional dependency

The Bioconductor package qvalue is used for q-value estimation when available. If qvalue is not installed, DANDELION falls back to Benjamini-Hochberg adjusted p-values.

# Optional
# install.packages("BiocManager")
BiocManager::install("qvalue")

Main functions

Function	Description
med_gene()	Applies DANDELION to identify significant exposure-gene2 mediation pairs. The exposure can be a distal gene or a SNP.
calc_pair.gene()	Organizes significant gene-based DANDELION results and merges nearby exposure genes into locus-level regions.
calc_pair.snp()	Organizes significant SNP-based DANDELION results and maps SNP exposures to cis genes.
gen_fig()	Generates a network visualization of DANDELION-prioritized gene-pair results.

Required input data

1. p.trans: trans-association p-value matrix

p.trans is a numeric matrix of trans-regulatory association p-values.

• Rows correspond to candidate disease-proximal genes (gene2).
• Columns correspond to exposures (gene1), either genes or SNPs.
• Row names and column names must be provided.

For gene-based DANDELION:

rows    = candidate proximal genes / trans target genes
columns = distal regulatory genes

For SNP-based DANDELION:

rows    = candidate proximal genes / trans target genes
columns = SNPs

2. p.wes: gene-level disease association p-values

p.wes is a named numeric vector of gene-level disease association p-values.

Examples include:

• WES burden-test p-values,
• gene-level GWAS p-values,
• other gene-level association statistics.

The names of p.wes should match the row names of p.trans.

3. ref.table: gene annotation table

ref.table provides gene position and gene type information.

It must contain the following columns:

Column	Description
gene_name	Gene symbol or gene identifier
type	Gene type, such as protein_coding or lincRNA
Chromosome	Chromosome in chr1, chr2, ... format
start	Gene start position
end	Gene end position

DANDELION internally keeps autosomal protein_coding and lincRNA genes and removes genes on chrM, chrX, and chrY for the main trans-mediation analysis.

4. gene1.list: candidate exposure list

gene1.list is a character vector of candidate exposures to analyze.

• For gene-based DANDELION, it should contain candidate distal genes.
• For SNP-based DANDELION, it should contain candidate SNPs.
• Values must be present in colnames(p.trans).

5. SNP.ref: SNP annotation table for SNP-based DANDELION

When gene1.type = "SNP", SNP.ref must be provided.

It must contain:

Column	Description
SNP	SNP rsID
SNPPos	SNP genomic position
SNPChr	SNP chromosome, either 1 or chr1 format

6. uniq_snp: SNP-to-cis-gene mapping table

For calc_pair.snp(), uniq_snp is used to map SNP exposures to their corresponding cis genes.

It must contain:

Column	Description
SNP	SNP rsID
GeneSymbol	Corresponding cis gene symbol

Quick start: gene-based DANDELION

The following toy example demonstrates the gene-based workflow.

library(DANDELION)

set.seed(1)

# Example trans-association p-value matrix.
# Rows are candidate disease-proximal genes; columns are exposure genes.
p.trans <- matrix(runif(60, 0.001, 0.9), nrow = 12, ncol = 5)
rownames(p.trans) <- paste0("G", 1:12)
colnames(p.trans) <- paste0("E", 1:5)

# Example gene-level disease association p-values.
p.wes <- runif(12, 0.001, 0.9)
names(p.wes) <- paste0("G", 1:12)

# Example gene annotation table.
ref.table <- data.frame(
  gene_name = c(paste0("G", 1:12), paste0("E", 1:5)),
  type = "protein_coding",
  Chromosome = "chr1",
  start = seq_len(17) * 1e7,
  end = seq_len(17) * 1e7 + 1000
)

# Run gene-based DANDELION.
res <- med_gene(
  p.trans = p.trans,
  p.wes = p.wes,
  ref.table = ref.table,
  gene1.list = colnames(p.trans),
  gene1.type = "Gene",
  target.fdr = 0.1,
  dist = 5e6
)

res

The returned object is a structured dandelion_result object. You can extract information directly:

res$gene1
res$mat.sig[1:5, 1:3]
res$mat.p[1:5, 1:3]

Organize significant gene-gene pairs:

ref.table.keep <- ref.table[
  ref.table$type %in% c("lincRNA", "protein_coding") &
    !(ref.table$Chromosome %in% c("chrM", "chrX", "chrY")),
]

pair.res <- calc_pair.gene(
  mat.sig = res$mat.sig,
  mat.p = res$mat.p,
  p.wes = p.wes,
  gene1 = res$gene1,
  ref.table.keep = ref.table.keep,
  eta.wgs = 1e-5
)

pair.res

Quick start: SNP-based DANDELION

The following toy example demonstrates the SNP-based workflow.

library(DANDELION)

set.seed(1)

# Example SNP-to-gene trans-association p-value matrix.
# Rows are candidate disease-proximal genes; columns are SNP exposures.
p.trans <- matrix(runif(60, 0.001, 0.9), nrow = 12, ncol = 5)
rownames(p.trans) <- paste0("G", 1:12)
colnames(p.trans) <- paste0("rs", 1:5)

p.wes <- runif(12, 0.001, 0.9)
names(p.wes) <- paste0("G", 1:12)

ref.table <- data.frame(
  gene_name = paste0("G", 1:12),
  type = "protein_coding",
  Chromosome = "chr1",
  start = seq_len(12) * 1e7,
  end = seq_len(12) * 1e7 + 1000
)

SNP.ref <- data.frame(
  SNP = paste0("rs", 1:5),
  SNPPos = seq_len(5) * 2e7,
  SNPChr = "1"
)

uniq_snp <- data.frame(
  SNP = paste0("rs", 1:5),
  GeneSymbol = paste0("G", 1:5)
)

# Run SNP-based DANDELION.
res <- med_gene(
  p.trans = p.trans,
  p.wes = p.wes,
  ref.table = ref.table,
  gene1.list = colnames(p.trans),
  gene1.type = "SNP",
  SNP.ref = SNP.ref,
  target.fdr = 0.1,
  dist = 5e6
)

res

Organize significant SNP-gene pairs:

ref.table.keep <- ref.table[
  ref.table$type %in% c("lincRNA", "protein_coding") &
    !(ref.table$Chromosome %in% c("chrM", "chrX", "chrY")),
]

pair.res <- calc_pair.snp(
  mat.sig = res$mat.sig,
  mat.p = res$mat.p,
  p.wes = p.wes,
  gene1 = res$gene1,
  uniq_snp = uniq_snp,
  ref.table.keep = ref.table.keep,
  SNP.ref = SNP.ref,
  eta.wgs = 1e-5
)

pair.res

Output objects

Output from med_gene()

med_gene() returns an object of class dandelion_result.

Output	Description
gene1	Exposures with at least one valid DANDELION result
mat.sig	Matrix encoding significant gene1-gene2 pairs
mat.p	Matrix of DANDELION p-values
gene1.type	Exposure type, either Gene or SNP
target.fdr	FDR cutoff used for identifying significant pairs

Output from calc_pair.gene() and calc_pair.snp()

These functions return objects of class dandelion_pairs.

Output	Description
pairs_dact	Significant DANDELION pairs before locus-level merging
gene.pair	Pair table after organizing exposure genes into loci
sig_gene2	DPGs that are also significant in gene-level burden tests
non_sig.gene2	DPGs that are not significant in gene-level burden tests

Network visualization

gen_fig() generates a network visualization from organized DANDELION gene-pair results.

gen_fig(
  gene.pair = pair.res$gene.pair,
  p.wes = p.wes,
  eta.wgs = 1e-5,
  pic_dir = "path/to/output/folder"
)

The function saves Network.pdf to pic_dir and invisibly returns the output file path.

Verbose output

By default, DANDELION functions do not print progress messages to the console. To show progress messages, set verbose = TRUE:

res <- med_gene(
  p.trans = p.trans,
  p.wes = p.wes,
  ref.table = ref.table,
  gene1.list = colnames(p.trans),
  gene1.type = "Gene",
  verbose = TRUE
)

Analysis directory

The Analysis/ directory contains public templates and example scripts for running DANDELION simulation and real-data workflows.

During peer review, this repository provides lightweight, GitHub-friendly templates that demonstrate the expected input structure, DANDELION function calls, output format, and evaluation strategy. Full manuscript-scale data-generation scripts, internal HPC pipelines, real-data analysis scripts, and intermediate results are maintained privately during peer review to protect code provenance and avoid unauthorized redistribution.

Analysis/simulation/

This folder contains simulation scripts used to evaluate the statistical performance of DANDELION under controlled mediation settings.

Available scripts include:

• simulation_compare.R
  Implements the manuscript-scale simulation framework comparing DANDELION with ARCHIE. The simulation generates three independent samples to mimic realistic scenarios where genetic variants, gene expression, and complex traits are measured in non-overlapping cohorts. The framework evaluates gene-based DANDELION, SNP-based DANDELION, and ARCHIE in terms of statistical power and global false discovery rate (FDR) control.

• sim_global_qvalue.R
  Evaluates the global FDR calibration of DANDELION under controlled simulation settings. This simulation generates disease-distal gene to target gene mediation structures and assesses gene-level and pair-level discovery performance across prespecified FDR thresholds using q-value based inference.

• global_fdr_simulation_template.R
  Provides a simplified example demonstrating how to evaluate DANDELION at both the gene level and exposure-gene pair level, including performance metrics such as true positive proportion (TPP), FDR, and the number of selected genes/pairs.

• simulation_compare_template.R
  Provides a lightweight example illustrating how to organize simulation workflows comparing different DANDELION analysis strategies using shared input/output structures.

The simulation framework evaluates:

• Statistical power: the proportion of true disease-proximal genes correctly identified.
• False discovery rate control: the ability of DANDELION to maintain the prespecified FDR levels.
• Gene-level and pair-level performance: evaluation of disease-proximal gene discovery and exposure-gene pair prioritization.
• Method comparison: comparison of DANDELION with alternative approaches under matched simulation settings.

The manuscript-scale simulation implements the DANDELION testing framework and downstream inference procedures used in the study, while the template scripts provide simplified examples for applying DANDELION in user-defined simulation settings.

---

Analysis/real_data/

This folder contains scripts for applying DANDELION to real disease-trait datasets using SNP-based trans-regulatory summary statistics and gene-level disease association results.

Available scripts include:

• Analysis/real_data/run_dandelion_snp_pipeline.R
  A general SNP-based DANDELION workflow illustrating the application of DANDELION using trans-eQTL summary statistics, SNP annotation, SNP-to-gene mapping information, and trait-specific gene-level disease association results.

• Analysis/real_data/run_dandelion_asthma_pipeline.R
  The trait-specific SNP-based DANDELION workflow used for the asthma analysis reported in the manuscript.

• Analysis/real_data/run_dandelion_real_data.R
  Implements the generalized real-data analysis workflow used for multi-trait analyses in the manuscript. The same pipeline was applied across multiple disease traits by changing trait-specific input files.

The real-data workflows include:

• loading trans-regulatory eQTL summary statistics;
• preparing SNP annotation and SNP-to-gene mapping information;
• preparing trait-specific gene-level disease association p-values;
• applying med_gene() for trans-regulatory mediation analysis;
• performing downstream SNP-gene pair prioritization;
• saving DANDELION results, significant SNP-gene pairs, and summary outputs.

Citation

If you use DANDELION, please cite the following work.

The statistical testing method in DANDELION builds on the Divide-Aggregate Composite-null Test (DACT):

Liu, Z., Shen, J., Barfield, R., Schwartz, J., Baccarelli, A. A., & Lin, X. (2022). Large-Scale Hypothesis Testing for Causal Mediation Effects with Applications in Genome-wide Epigenetic Studies. Journal of the American Statistical Association, 117(537), 67–81. https://doi.org/10.1080/01621459.2021.1914634

The trans-gene regulation paper is here:

Salamone, I.M.†, Tian, P.†, Qi, Z., Zhao, J., Zhang, L., Tan, Q., Li, J., Michael, A.N., Thornburg, A.G., Sakabe, N.J. and Minogue, M., 2026. Trans-regulatory gene mapping prioritizes disease drivers in asthma. Cell. https://www.cell.com/cell/fulltext/S0092-8674(26)00866-4

Note: † indicates authors contributed equally to this work.

Authorship and copyright

R implementation: Peixin Tian

Copyright: Copyright (C) 2025 Peixin Tian.

License: GPL-3.0. See the LICENSE file.

License

This package is distributed under the GNU General Public License v3.0. See the LICENSE file for details.