networkR

networkR is an R package scaffold for configurable Bayesian network analysis in factorial and repeated-measures studies.

I intend to extend this to gaussian copulas/other graphical models, but for now, just Bayesian networks.

Current scope

• YAML or programmatic configuration
• wide_by_tissue and long-format panel data ingestion, with automatic measurement column inference
• Derived design variables
• MICE predictive mean matching imputation
• Hartemink or factor-based discretization with fallback handling
• Tier-aware blacklist construction
• Bootstrap Bayesian network fitting — stagedtrees backend (default) and bnlearn backend
• ggraph(layout = "fr") network plotting; event tree / CEG visualization for the stagedtrees backend
• MCP server for natural-language-driven configuration and pipeline execution

Quick start — MCP-driven workflow (recommended)

The package ships with an MCP server at inst/mcp_server.R. Open the repository in VS Code; the server starts automatically from .vscode/mcp.json.

# In GitHub Copilot Chat (or any MCP-compatible client)

1. "Explore my data directory at data/"
   → networkr_explore_data_directory reports column roles and a reviewScaffold per file.

2. "Generate a config from data/subjectIdTable_TB.csv"
   → networkr_generate_configuration_from_data returns a YAML draft and a columnRoles scaffold.

3. Review the scaffold. Move columns between roles as needed, then:
   "Refine the config using this corrected scaffold: <paste edited columnRoles block>"
   → networkr_refine_configuration_from_scaffold regenerates YAML from the corrected assignments.

4. "Save the config to configs/my_analysis.yml"
   → networkr_write_configuration_file writes and immediately validates the file.

5. "Run the pipeline from configs/my_analysis.yml"
   → networkr_run_pipeline_from_config runs the full analysis.

6. "Describe the results"
   → networkr_describe_last_pipeline_result summarizes the most recent run.

See vignettes/networkR-mcp.Rmd for a full step-by-step walkthrough.

Quick start — programmatic

library(networkR)

configuration <- ParseConfiguration("configs/my_analysis.yml")
analysisResult <- RunPipeline(configuration)

analysisResult$plotObjects$fullModel

configs/lung_ldln_bidirectional_template.yml is an annotated reference config for wide_by_tissue bidirectional analyses; copy and edit the column-name fields to match your data.

Status

This repository contains the package foundation. Remaining work includes joint paired-tissue modeling, mediation workflows, and correlation reporting.