MR-CCC: Bayesian Mendelian Randomization for Causal Cell-Cell Communication

Code for the paper:

Bayesian Mendelian Randomization for Causal Cell-Cell Communication
Bitan Sarkar and Yang Ni, 2026

---

Repository Contents

mr_ccc_gibbs.cpp

The core Gibbs sampler implementing the MR-CCC model. Written in C++ using RcppArmadillo. Compile in R with:

Rcpp::sourceCpp("mr_ccc_gibbs.cpp")

This makes the function mr_ccc_gibbs() available in your R session.

build_lr_database.R

Builds the ligand-receptor (LR) pair database by combining:

• CellPhoneDB v5 — curated LR interactions with gene symbols and pathway annotations. Download the required input files from: https://github.com/ventolab/cellphonedb-data/tree/master/data
• MSigDB Reactome (C2/CP:REACTOME) — pathway gene sets, accessed via the msigdbr R package (no download needed).

Set DATA_DIR at the top of the script to the folder containing the CellPhoneDB CSV files, then source() the script. Produces two objects: lr_matrix and key_cols.

real_data_analysis.R

Runs the MR-CCC analysis for one ordered sender → receiver cell-type pair, iterating over all ligand-receptor-pathway triplets. Also produces three publication figures (bubble plot, PIP lollipop, effect curves).

To change the cell-type pair, edit the two lines at the top of the USER SETTINGS block:

Cell1 <- "NKCells"         # sender cell type
Cell2 <- "MonocytesCells"  # receiver cell type

Valid values for both Cell1 and Cell2:

Value	Cell type
"BCells"	B cells
"CD4Cells"	CD4⁺ T cells
"CD8Cells"	CD8⁺ T cells
"NKCells"	NK cells
"MonocytesCells"	Monocytes

Any ordered combination is valid, e.g. Cell1 = "BCells", Cell2 = "CD4Cells" runs the B cells → CD4⁺ T cells analysis.

Prerequisites before sourcing:

Rcpp::sourceCpp("mr_ccc_gibbs.cpp")
source("build_lr_database.R")

simulation_mrccc.R

Simulation study comparing four methods (OLS, MVMR, MR-BMA, MR-CCC) across three scenarios and four sample sizes. Produces summary tables and boxplot figures.

Prerequisites before sourcing:

Rcpp::sourceCpp("mr_ccc_gibbs.cpp")

---

Data

Preprocessed OneK1K single-cell RNA-seq and donor data used in the real-data analysis are available at:

https://doi.org/10.5281/zenodo.19675075

Download B_T_NK_monocytes.rda and donor.rda, save them to a local folder, and set DATA_DIR at the top of real_data_analysis.R to that folder path.

---

Dependencies

# CRAN
install.packages(c(
  "Rcpp", "RcppArmadillo",
  "dplyr", "tidyr", "ggplot2", "forcats", "scales",
  "purrr", "tibble", "stringr", "ggrepel", "ggtext",
  "msigdbr", "parallel"
))

# Bioconductor
BiocManager::install(c("AUCell", "UCell", "GSVA", "GenomicRanges"))

A C++ compiler compatible with C++11 is required (Xcode Command Line Tools on macOS; Rtools on Windows).

---

License

MIT License. See LICENSE for details.