Use case

3. Use case#

See Use Case section for how to start an analysis task.

3.1. Automating Configuration with LLMs#

Manually writing sample_sheet.csv and project_config.yaml is tedious and error-prone, especially for large cohorts with complex naming conventions. The YULUMINA repository provides a clindet skill that lets you configure a Clindet analysis entirely through conversation with an LLM — no hand-editing of config files required.

3.1.1. How it works#

The clindet skill encodes Clindet’s configuration rules, software profiles, run-type matrix, and sample-naming conventions into structured reference files. When loaded by an LLM (Claude Code, OpenAI Codex CLI, or any tool that supports skills), it conducts an interview with you to collect:

  • Sequencing type (WES / WGS / targeted panel / RNA-seq)

  • Species and genome version

  • FASTQ file locations and naming patterns

  • BED file information

  • Project name, output directory, and analysis scope

  • Compute environment (local node or Slurm HPC)

After the interview, the LLM generates all required project files:

  • project_config.yaml — workflow configuration

  • sample_sheet.csv — sample metadata and pairing

  • run_dryrun.sh — dry-run command for validation

  • submit_pipeline.sh — production submission script

  • run_info.md — run summary for documentation

  • data_sanity_report.md — data quality notes for review

3.1.2. Installation#

Clone YULUMINA and symlink the clindet skill into your LLM tool’s skills directory:

Claude Code:

git clone https://github.com/zyllifeworld/YULUMINA.git /your/preferred/path/YULUMINA
ln -s /your/preferred/path/YULUMINA/clindet ~/.claude/skills/clindet

OpenAI Codex CLI:

git clone https://github.com/zyllifeworld/YULUMINA.git /your/preferred/path/YULUMINA
ln -s /your/preferred/path/YULUMINA/clindet ~/.codex/skills/clindet

3.1.3. Usage#

Once the skill is installed, start a conversation with your LLM and describe your analysis. For example:

“I have paired WES data for 10 tumor-normal samples under ~/projects/cancer_wes/data/. The files are named like T__R1.fq.gz. I want to run somatic SNV and CNV calling on b37.”

The LLM will ask clarifying questions, then generate all config files ready to use with the snakemake commands described in Run the Quick Test.

For details on the skill’s capabilities and design, see the YULUMINA README.