Article_ECOTOX_protocol

Computational protocol repository for the PPDB-centred ECOTOX manuscript. This repository owns the calculations, local data snapshots, figures, and machine-readable results used by the manuscript repository.

The manuscript text lives in the sibling repository:

parent-folder/
  Article_ECOTOX_protocol/
  Article_ECOTOX_manuscript/

The manuscript reads figures and generated tables from this repository through relative paths such as ../Article_ECOTOX_protocol/figures/... and ../Article_ECOTOX_protocol/results/generated_tables/.... Keep both repositories as siblings with these directory names.

Repository Rule

Whenever a calculation, input, output location, or manuscript-facing result changes, update the relevant README in the same change. The README files are part of the reproducibility contract, not optional notes.

Layout

• code/: helper code and small static input tables used by the pipeline.
• data/: repository-local input snapshots, including optional PRISTIN toxic-unit (TU) imports when the OneDrive source files are available.
• notebooks/: executable notebooks and scripts for database building, feature audit, clustering, quantitative structure-property relationship (QSPR) benchmarking, and supervised TU modelling.
• results/: generated CSV, H5AD, JSON, and LaTeX table outputs consumed by the manuscript.
• figures/: generated manuscript and supplementary figures.
• reports/: auxiliary text outputs and diagnostics.
• .codex/: traceability folder with user queries, helper scripts, and work history.

Main Commands

Build the local PPDB parent database and QSPR targets:

make database

Copy optional PRISTIN TU source data into repository-local CSV snapshots if the OneDrive files exist:

make update-data

Audit feature redundancy, PCA explained variance, and missingness classes:

make feature-audit

Run exploratory supervised PRISTIN TU models:

make tu-models

Run the full analysis notebook and regenerate figures/results:

make analysis

Regenerate LaTeX table fragments consumed by the manuscript:

make tables

Run the standard calculation set:

make all

Current Outputs Used by the Manuscript

The manuscript expects these protocol outputs to exist:

• figures/joint_pca_scatter.png: principal component analysis (PCA) scatter with axis labels including explained variance.
• figures/pristin_tu_model_r2.png: exploratory TU model cross-validated R2 summary.
• figures/pristin_tu_feature_correlations.png: exploratory feature-TU Spearman correlation summary.
• results/generated_tables/*.tex: LaTeX table fragments included by the supplementary information.
• results/pristin_tu_model_summary.csv
• results/pristin_tu_feature_correlations.csv
• results/pristin_tu_predictions.csv
• results/feature_redundancy_audit.csv
• results/feature_missingness_audit.csv
• results/feature_pca_variance_audit.csv

PRISTIN TU Inputs

The optional PRISTIN import looks for:

/Users/jordivilla/Library/CloudStorage/OneDrive-SharedLibraries-UniversitatdeVic/PRISTIN-Equip - General/RESULTS WP3 DANI/PRISTIN_pes_water_l_NETA.xlsx
/Users/jordivilla/Library/CloudStorage/OneDrive-SharedLibraries-UniversitatdeVic/PRISTIN-Equip - General/RESULTS WP3 DANI/PRISTIN_pes_water_l_c_pnec_tu-2.xlsx

If present, make update-data writes repository-local CSV snapshots under data/. If absent, the rest of the repository remains usable with the checked-in or previously generated local snapshots.

Current TU Model Status

The current primary PRISTIN TU source provides 37 compounds, of which 25 match the PPDB parent database by normalized names. Partial least squares (PLS), Random Forest, and support-vector regression (SVR) models are run as a supervised training layer, but all current cross-validated R2 values are negative. Treat the TU layer as a reproducible scaffold and feature-prioritization analysis until compound matching and TU coverage improve.

Build Order With Manuscript

From the parent folder:

cd Article_ECOTOX_protocol
make all

cd ../Article_ECOTOX_manuscript
make