Source: https://github.com/aipoch/medical-research-skills

lncRNA Regulatory Network Construction Analysis

When to Use

Use this skill when the user wants a local-database network lookup workflow rather than expression-based inference.

Typical use cases:

• Build an lncRNA-mRNA network from target genes and the bundled ceRNA reference tables
• Start from a candidate lncRNA list and retrieve linked mRNAs through shared miRNAs
• Generate an auditable lncRNA-mRNA network table plus a tripartite evidence table
• Reuse a saved database-derived network object to regenerate a PDF plot

Do not use this skill when the user asks for:

• Expression-matrix-based network inference
• Correlation analysis between lncRNAs and mRNAs
• Causal inference or regulatory-strength estimation from expression data
• Online database querying or remote API lookups

Execution Model

This is a hybrid skill.

1. Read SKILL.md to confirm that the request is database-driven.
2. Use scripts/main.R for actual execution.
3. Use --mode analyze to build tables and a saved .rda object.
4. Use --mode visualize to reuse the saved object and redraw the PDF without rebuilding the database tables.
5. Use --mode full to run both steps in one pass.
6. Read reference files only when more detail is needed.
7. Before --mode visualize, confirm that output_dir/data/lncrna_network.rda already exists.
8. In visualize mode, the saved .rda object is the required input; a missing or invalid reference_dir does not block plot reuse.
9. After execution, report the mode, output directory, key files, and either the retained network size or the surfaced skill error code.

When to Read External Files

Out-of-Scope Response Pattern

If the request is expression-based rather than database-driven, do not run this skill. Respond briefly with:

This skill only projects lncRNA-mRNA links from local ceRNA reference tables using target gene and/or lncRNA lists. It does not infer networks from expression matrices or estimate causal regulatory strength. Use a different workflow for expression-based correlation or causal inference.

If the request is ambiguous between database-driven lookup and expression-based inference, ask one short clarifying question before running any command.

Agent Response Contract

For a successful run, report:

• The selected mode and why it fits the request
• The output_dir
• The key output files that were generated or reused
• The retained network size from table/network_stats.txt when available
• A short reminder that the result is database-driven rather than expression-inferred

For a failed run, report:

• The surfaced SKILL_* error code
• The most likely cause based on references/troubleshooting.md
• The shortest actionable next step for rerunning the workflow

Usage

Rscript scripts/main.R \
  --mode full \
  --target_genes ./target_genes.txt \
  --target_lncrna ./target_lncrna.txt \
  --mirna_dataset combined \
  --lncrna_strictness High \
  --min_shared_mirna 1 \
  --reference_dir ./references/database \
  --output_dir ./output \
  --seed 42

Arguments

Input Format

Target Gene List

• Plain-text file or comma-separated list
• One gene symbol per line when using a file

Example:

TP53
BRCA1
MYC

Target lncRNA List

• Plain-text file or comma-separated list
• One lncRNA symbol per line when using a file

Example:

XIST
SNHG16
HNRNPU-AS1

At least one of --target_genes or --target_lncrna must be provided.

Output Files

Error Handling

Progressive Disclosure

1. Start with --target_genes or --target_lncrna.
2. Add the second target list if a more focused subnetwork is needed.
3. Switch --mirna_dataset if a different miRNA-mRNA evidence source is required.
4. Adjust --lncrna_strictness, --lncrna_freq_thresh, and --min_shared_mirna to tighten or relax the projected network.
5. Reuse --mode visualize once the .rda object exists.

Result Size Guidance

• Broad gene-only or lncRNA-only runs can expand quickly and may retain hundreds to thousands of edges.
• If the retained network is too large for practical review, report the edge and node totals, then increase --min_shared_mirna, increase --lncrna_freq_thresh, or provide the complementary target list.
• Start with the bundled demo inputs before moving to broader target lists.

Examples

Gene-Driven Network

Rscript scripts/main.R \
  --mode full \
  --target_genes ./target_genes.txt \
  --reference_dir ./references/database \
  --output_dir ./output

lncRNA-Driven Network

Rscript scripts/main.R \
  --mode analyze \
  --target_lncrna ./target_lncrna.txt \
  --mirna_dataset starbase \
  --lncrna_strictness Median \
  --output_dir ./lncrna_only_output

Focused Bipartite Network

Rscript scripts/main.R \
  --mode full \
  --target_genes TP53,BRCA1,MYC \
  --target_lncrna XIST,SNHG16,HNRNPU-AS1 \
  --mirna_dataset combined \
  --lncrna_strictness High \
  --min_shared_mirna 2 \
  --output_dir ./focused_output

Visualization Reuse

Rscript scripts/main.R \
  --mode visualize \
  --output_dir ./focused_output \
  --plot_file reused_network.pdf \
  --layout_type fr

For the bundled baseline and CLI notes, read references/cli-guide.md.

Testing

Rscript scripts/main.R --help

Rscript tests/run_tests.R

Rscript scripts/main.R \
  --mode full \
  --target_genes tests/data/target_genes.txt \
  --target_lncrna tests/data/target_lncrna.txt \
  --reference_dir references/database \
  --output_dir tests/output \
  --seed 42

Expected retained outputs after a validated run:

• tests/output/table/lncrna_mrna_edges.csv
• tests/output/table/lncrna_mirna_mrna_evidence.csv
• tests/output/table/lncrna_mrna_nodes.csv
• tests/output/table/network_stats.txt
• tests/output/data/lncrna_network.rda
• tests/output/plot/lncrna_mrna_network.pdf
• tests/output/session_info.txt
• tests/output/output_manifest.txt
• tests/output/run_record.txt

Scope Limits

This skill does not infer networks from expression matrices and does not perform online queries.

If the user needs expression-based correlation or causal inference, use a different workflow.