powidla/Friend-Or-Foe

Overview

Welcome to the Friend or Foe Collection!

Friend or Foe is a collection of synthetic environmental tabular datasets generated by solving metabolic models for 10,000 pairs of bacteria.

This README contains the description of the Friend or Foe collection of bacterial datasets. The general tree structure of the repository is described by Friend-or-Foe Structure figure.

Friend or Foe Collection (panel a) contains data suitable for 5 different ML methods stored in corresponding folders (panel b): Classification, Regression, Transfer Learning, Clustering and Generative modeling.

Each folder contains data obtained from two sources of metabolic models, AGORA and CARVEME (panel c). For each source there is 100 and 50 additional compounds used in the metabolic modeling and the data corresponds to each folder - panel d.

Next on the panel e, the datasets are named according to the Table 3 in the manuscript, i.e Binary Classification for Facultative Cooperation vs Competition is BC-I.

Finally, each subfolder contains .csv train/test files. Note that for Clustering task data is stored as raw .npy arrays for compilatibility with clustering algorightms (see manuscript).

In addition to tabular data, we provide useful information regarding the metab models and species (panel g). Specifically, General Info contains information about organism names and abbreviations of environmental compounds (to convert the abbreviation to a name, see Agora_env_compounds.xlsx for AGORA and the Bigg database at http://bigg.ucsd.edu for CARVEME).

MetabModels stores used metabolic models in .mat format for both AGORA and CARVEME. On top of that, we provide raw results of metabolic modeling in raw subfolder (panel h).

Finally, we store reference weights of ML models for benchmarks in Models subfolder.

Tabular structure

First 424 columns for AGORA and 499 for CARVEME identify the abreviations for chemical compounds. The values are binary 0 or 1, identifying the presense or absence of a particular compound in the environment.

The last column is the target, for classification is an integer, e.g num of class; while for regression the target is growth rate.

Overview of the dataset compendium

The compendium comprises 64 datasets, each corresponding to a distinct machine learning Task. The Samples column indicates the number of environments used per task, with additional 100 or 50 compounds added to enhance diversity. The Group denotes the number of additional compounds, while Collection specifies whether the dataset belongs to AGORA or CARVEME. AGORA datasets contain 424 unique chemical features, whereas CARVEME datasets contain 499.

Task	AGORA (100)	AGORA (50)	CARVEME (100)	CARVEME (50)	Metric
BC-I	326,331	89,500	332,938	167,918	Acc / MCC
BC-II	624,743	547,175	399,751	466,814	Acc / MCC
BC-III	509,298	491,167	478,057	486,646	Acc / MCC
BC-IV	339,006	198,485	237,535	211,179	Acc / MCC
BC-V	223,561	142,477	315,841	231,011	Acc / MCC
MC-I	730,186	563,922	605,373	560,689	Acc / MCC
MC-II	925,828	1,074,570	655,089	922,678	Acc / MCC
MC-III	1,252,159	1,164,072	988,027	1,090,596	Acc / MCC
GR-I	105,443	16,747	339,004	93,875	RMSE / r²
GR-II	220,888	72,755	205,622	93,875	RMSE / r²
GR-III	403,855	474,420	272,435	392,771	RMSE / r²
TL-I	659,269	257,420	494,249	454,572	Acc / MCC
TL-II	1,997,248	1,580,917	1,848,506	1,729,659	Acc / MCC
US-I	12,002	12,000	9,004	8,000	DCSI / SC
US-II	12,001	3,000	11,000	8,000	DCSI / SC
GEN	40,000	20,000	40,000	20,000	α-Precision / β-Recall

Getting started

from huggingface_hub import hf_hub_download
import pandas as pd

REPO_ID = "powidla/Friend-Or-Foe"
X_train_ID = "Classification/AGORA/100/BC-I/X_train_BC-I-100.csv"
X_val_ID = "Classification/AGORA/100/BC-I/X_val_BC-I-100.csv"
X_test_ID = "Classification/AGORA/100/BC-I/X_test_BC-I-100.csv"

y_train_ID = "Classification/AGORA/100/BC-I/y_train_BC-I-100.csv"
y_val_ID = "Classification/AGORA/100/BC-I/y_val_BC-I-100.csv"
y_test_ID = "Classification/AGORA/100/BC-I/y_test_BC-I-100.csv"

X_train = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=X_train_ID, repo_type="dataset"))
X_val = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=X_val_ID, repo_type="dataset"))
X_test = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=X_test_ID, repo_type="dataset"))

y_train = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=y_train_ID, repo_type="dataset"))
y_val = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=y_val_ID, repo_type="dataset"))
y_test = pd.read_csv(hf_hub_download(repo_id=REPO_ID, filename=y_test_ID, repo_type="dataset"))

A quick-start example is available in our github repository.

Citation Information

@ARTICLE{Solowiej-Wedderburn2025-ar,
  title     = "Competition and cooperation: The plasticity of bacterial
               interactions across environments",
  author    = "Solowiej-Wedderburn, Josephine and Pentz, Jennifer T and Lizana,
               Ludvig and Schroeder, Bjoern O and Lind, Peter A and Libby, Eric",
  journal   = "PLoS Comput. Biol.",
  publisher = "Public Library of Science (PLoS)",
  volume    =  21,
  number    =  7,
  pages     = "e1013213",
  month     =  jul,
  year      =  2025,
  copyright = "http://creativecommons.org/licenses/by/4.0/",
  language  = "en"
}

@misc{cherednichenko2025friendfoe,
      title={Friend or Foe}, 
      author={Oleksandr Cherednichenko and Josephine Solowiej-Wedderburn and Laura M. Carroll and Eric Libby},
      year={2025},
      eprint={2509.00123},
      archivePrefix={arXiv},
      primaryClass={q-bio.QM},
      url={https://arxiv.org/abs/2509.00123}, 
}