corr-bench

Plotting code for the figures of "Quality, not quantity, of enriched pathways should guide benchmarks of batch correction methods in biological networks" (Babaie, Quackenbush, Ben Guebila).

This repository introduces 2 new COBRA modes (available in netzoopy 0.12):

• Nonnegative Least squares (NNLS)
• Nonnegative Lasso (NNLasso)

In addition to the default MLE mode. These 2 modes come as an improvement to the MLE mode which tends to produce between 0-5% of |values|>1 in the predicted co-expression matrices (Table S1 of the COBRA publication). By constraining estimated matrices to be positive-semi definite, NNLS and NNLasso produce correctly scaled correlation coefficients. The differences are small in practice as reflected in figure 4, reflecting that correlation violations are rare.

Specifically, this repository contains the final plotting scripts for the benchmarks and their output figures. Analysis results (per-tissue WGCNA modules, ORA/GSEA tables, COBRA train/test R², runtime benchmarks, GMT files) are released on Zenodo (10.5281/zenodo.20187237) and should be placed under results/ and data/ matching the paths below.

Gene filtering note

All COBRA and CorrAdjust results in results/tpmfilter_data/, results/wgcna_ora_cov_all/, and results/wgcna_ora_hardy/ use the same gene set per tissue: genes with median TPM ≥ 1 across the training samples. Benchmarks with median raw count ≥ 1 were used to obtain a larger gene set than TPM ≥ 1 for runtime comparison.

Data (Zenodo)

The deposit contains 5 tarballs and 2 TSVs. After download, extract each tarball at the indicated location:

Archive	Extract at	Used by
tpmfilter_data.tar.gz (~2.2 GB)	results/ → results/tpmfilter_data/	Figs 1, 2, 3, S1, S2, S3
wgcna_ora_cov_all.tar.gz (~19 MB)	results/ → results/wgcna_ora_cov_all/	Figs 2, 3, S3
wgcna_ora_hardy.tar.gz (~13 MB)	results/ → results/wgcna_ora_hardy/	Fig 2 (Hardy tissues)
GMT_files.tar.gz (~2.2 MB)	data/input_data/ → data/input_data/GMT_files/	Figs S1, S2
ora_top100_all.tar.gz (~1.9 MB)	figures/misc/ → figures/misc/ora_top100_all/	Figs S1, S2
cobra_train_test_r2.tsv (4 KB)	results/	Fig 4
benchmark_medcount_summary.tsv (4 KB)	results/	Fig 3 (panel B, runtime)

Quick extract (run from repo root):

tar -xzf tpmfilter_data.tar.gz       -C results/
tar -xzf wgcna_ora_cov_all.tar.gz    -C results/
tar -xzf wgcna_ora_hardy.tar.gz      -C results/
tar -xzf GMT_files.tar.gz            -C data/input_data/
mkdir -p figures/misc && tar -xzf ora_top100_all.tar.gz -C figures/misc/
mv cobra_train_test_r2.tsv benchmark_medcount_summary.tsv results/

Figures

Figure	Script	Generates
Fig 1	src/fig1_reproduce_5b_bottom.py	figures/fig1_ges_gtex.png
Fig 2	src/plot_fig2_module_stacked.py	figures/fig2_module_stacked.png
Fig 3A	src/plot_3bar_precision.py	figures/fig_3bar_precision.png (intermediate)
Fig 3	src/plot_fig3_combined.py	figures/fig3_combined.png
Fig 4	src/18_plot_cobra_r2.py	figures/fig4_cobra_train_test_r2.png
Fig S1	src/plot_figs1_figs2.py	figures/figs1_ora_cp.png
Fig S2	src/plot_figs1_figs2.py	figures/figs2_gsea_cp.png
Fig S3	src/plot_fig3_summary_combined.py	figures/figs3_summary_gobp_combined.png

Expected data layout

After extracting the Zenodo tarballs (see Data (Zenodo)), files sit at:

results/
  benchmark_medcount_summary.tsv             # fig 3 (panel B)
  cobra_train_test_r2.tsv                    # fig 4
  wgcna_ora_cov_all/<tissue>/MLE/{module_assignments,ora_gobp,ora_cp}.tsv
  wgcna_ora_hardy/<tissue>/MLE/{module_assignments,ora_gobp,ora_cp}.tsv
  tpmfilter_data/
    wgcna_ora/<tissue>/CorrAdjust/{module_assignments,ora_gobp,ora_cp}.tsv
    gsea_all/<tissue>/...
    corradjust_gtex_outputs_models/<tissue>/...

data/
  input_data/GMT_files/c2.cp.v2023.2.Hs.symbols.gmt

<tissue> is one of the 11 GTEx tissues used in the paper (e.g. GTEx-Adipose-Subcutaneous, GTEx-Nerve-Tibial).

Install

pip install -r requirements.txt

Reproduce

python src/fig1_reproduce_5b_bottom.py
python src/plot_fig2_module_stacked.py
python src/plot_3bar_precision.py        # produces fig 3 panel A
python src/plot_fig3_combined.py         # combines panel A + runtime panel B
python src/18_plot_cobra_r2.py
python src/plot_figs1_figs2.py
python src/plot_fig3_summary_combined.py

License

MIT — see LICENSE.