WiNN Technical Correction Companion Repository

Companion code repository for the manuscript WiNN enables selective correction of run-order drift and batch effects while preserving biological signal in metabolomics data.

Author: Tanmay Tanna (ttanna@ethz.ch)

Contents

• manuscript/winn_manuscript.pdf: manuscript PDF.
• data/simulated/: simulated benchmark data.
• data/public/: public MTBLS79 input/output locations.
• notebooks/simulation_comparison.Rmd: simulated benchmark analysis.
• notebooks/public_data_comparison.Rmd: public MTBLS79 benchmark analysis.
• notebooks/rendered/: rendered notebook outputs.
• scripts/: data-generation and public-data preprocessing utilities.

Data Preparation

Populate the data directories before rendering the notebooks.

Simulated benchmark

Rscript scripts/simulate_metabolomics_benchmark.R data/simulated

Public MTBLS79 benchmark

1. Download Dataset07__SFPM.xlsx from the MetaboLights MTBLS79 record.
2. Save it at data/public/raw/Dataset07__SFPM.xlsx.
3. Preprocess it with:

Rscript scripts/preprocess_mtbls79_public_data.R \
  data/public/raw/Dataset07__SFPM.xlsx \
  data/public/processed

Analysis

Run the analyses through the notebooks after the data are available. The notebooks handle dependency installation for the comparison methods, perform subset-based tuning where applicable, and then execute full-data correction and evaluation.

Rscript -e "rmarkdown::render('notebooks/simulation_comparison.Rmd', output_dir = 'notebooks/rendered')"
Rscript -e "rmarkdown::render('notebooks/public_data_comparison.Rmd', output_dir = 'notebooks/rendered')"

TIGER and SERRF are the main runtime bottlenecks.