Metabolomic profiling of cassava and cocoyam during fermentation

This repository contains the code and analysis pipeline for an LC-MS/MS metabolomics study comparing cassava (Manihot esculenta) and cocoyam (Colocasia esculenta) tubers under control (unfermented) and fermented conditions. Three biological replicates were analyzed per condition.

Repository contents

notebooks/                           Analysis and figure generation
├── process_data_cassava.ipynb       Data integration pipeline (cassava)
├── process_data_cocoyam.ipynb       Data integration pipeline (cocoyam)
├── figure1.ipynb                    PCA and volcano plots
├── figure2.ipynb                    Chemical class distributions (violin and lollipop plots)
├── figure3.ipynb                    UpSet plots and annotation overlap
├── figure5.ipynb                    Pathway enrichment bubble plots (Mummichog)
└── supplemental_figures.ipynb       Cyanogenic glycoside boxplots and NPC significance

scripts/
├── biotransformer_sirius.py               BioTransformer parallel processing
└── compute_class_significance.py    Class-level Mann-Whitney U + BH-FDR (NPC and ClassyFire)

data/
├── processed_data/
│   ├── b3_cassavaonly/              SIRIUS/CANOPUS/GNPS outputs (Round 1, pre-BioTransformer) (cassava)
│   ├── b3_cassavaonly_biot/         SIRIUS/CANOPUS/GNPS outputs (Round 2, post-BioTransformer) + Mummichog results (cassava)
│   ├── b3_cocoyamonly/              SIRIUS/CANOPUS/GNPS outputs (Round 1, pre-BioTransformer) (cocoyam)
│   └── b3_cocoyamonly_biot/         SIRIUS/CANOPUS/GNPS outputs (Round 2, post-BioTransformer) + Mummichog results (cocoyam)
└── biotransformer/                  Compound lists used as input to BioTransformer

results/                             Generated by the notebooks (not deposited)
figures/                             Publication-ready figures (PNG and SVG, not deposited)

Figure 4 (molecular network visualization) was generated in Cytoscape and is not reproduced by this code.

Data availability

The processed input data is bundled with this repository under data/:

• data/processed_data/b3_cassavaonly_biot/ and data/processed_data/b3_cocoyamonly_biot/ — Round 2 (post-BioTransformer) SIRIUS/CANOPUS/GNPS outputs plus Mummichog results. All figures and downstream statistics in this repository run against these Round 2 files.
• data/processed_data/b3_cassavaonly/ and data/processed_data/b3_cocoyamonly/ — Round 1 (pre-BioTransformer) SIRIUS/CANOPUS/GNPS outputs, provided solely for reproducibility so a user who wants to re-run the upstream BioTransformer annotation loop can do so from scratch. Reproducing the figures does not require touching the Round 1 data.
• data/biotransformer/ — the compound lists used as input to BioTransformer.

Raw LC-MS/MS files and upstream MZmine feature-detection outputs are not included in this repository; deposition details for those will be added prior to publication. Files in results/ and figures/ are generated by running the notebooks and scripts and are not committed.

Reproducing the analysis

Requires Python 3.12 or later and Java (for BioTransformer).

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

1. Data integration

Run process_data_cassava.ipynb and process_data_cocoyam.ipynb. These merge the deposited SIRIUS/CANOPUS/GNPS annotations with quantified features, normalize by sample weight, impute missing values, and compute FDR-corrected differential abundance statistics. The notebooks read from the Round 2 (post-BioTransformer) directories data/processed_data/b3_*_biot/ — these are the final annotations used by every figure and statistical analysis in the paper. Outputs are written to results/.

The remainder of this section is only relevant if you want to re-run the upstream BioTransformer annotation loop from scratch. It is not required to reproduce any figure. The Round 1 (pre-BioTransformer) SIRIUS/CANOPUS/GNPS outputs deposited in data/processed_data/b3_*only/ are the inputs for this loop:

1. Point the process_data notebooks at the Round 1 directories and run them to generate data/biotransformer/b3_{batch}_compounds_for_biotransformer.csv.
2. Download BioTransformer 3.0 from https://bitbucket.org/wishartlab/biotransformer3.0jar and run:

python scripts/biotransformer_sirius.py --batch b3_cassavaonly --bt-dir /path/to/biotransformer
python scripts/biotransformer_sirius.py --batch b3_cocoyamonly --bt-dir /path/to/biotransformer

3. Load the predicted SMILES from results/biotransformer/{batch}/biotransformer_3step_predictions.csv into SIRIUS as a custom structure database, re-run SIRIUS/CSI:FingerID, and place the new outputs in data/processed_data/{batch}_biot/. Re-run the process_data notebooks to produce the final annotations. Features matched to BioTransformer-predicted structures appear with names prefixed BT_env_step* or BT_hgut_step*.

2. Class-level significance

python scripts/compute_class_significance.py

This computes Mann-Whitney U + BH-FDR for every NPC and ClassyFire chemical class across both tubers and writes:

• results/canopus_npc_significance.csv
• results/classyfire_significance.csv

These CSVs are consumed by figure2.ipynb and supplemental_figures.ipynb.

3. Figure generation

Run figure1.ipynb through figure5.ipynb, then supplemental_figures.ipynb.

Notebooks expect data files in data/ and write outputs to results/ and figures/separate_treatment/.

External tools

The following tools were used upstream of this repository to generate the input data:

Tool	Purpose
SIRIUS / CANOPUS	Molecular formula prediction, structure annotation, chemical classification
GNPS	Spectral library matching + Feature Based Molecular Networking
Mummichog	Pathway enrichment from m/z feature lists
BioTransformer 3.0	Metabolic transformation prediction
Cytoscape	Molecular network visualization (Figure 4)

Citation

Citation details will be added upon publication.