Citation

If you use NeKo in your research, please cite our paper:

Ruscone M, Tsirvouli E, Checcoli A, Turei D, Barillot E, et al. (2025) NeKo: A tool for automatic network construction from prior knowledge. PLOS Computational Biology 21(9): e1013300. https://doi.org/10.1371/journal.pcbi.1013300

Features

• Network creation and manipulation

• Connection of nodes and subnetworks

• Gene-to-phenotype mapping

• Network visualization and export helpers

• Interaction database integration

• Branching network history with automatic snapshots, HTML/SVG rendering, and state pruning controls

Connection strategies and topology

NeKo connection methods answer different biological and topological questions. Choose a strategy according to the intended flow of information rather than only according to runtime.

The diagram-based strategy guide illustrates every public connection strategy with fictitious nodes and includes a research-question decision table.

complete_connection separates path selection from reuse of topology already constructed during the same call:

maxlen is a mandatory positive edge cutoff for the new policies. For shortest-path policies it limits the search but does not request a longer path: if a two-edge path exists and maxlen=5, NeKo selects the two-edge shortest path or paths. all_bounded can be combinatorial as the cutoff grows.

The recommended explicit syntax is:

net.complete_connection(
    maxlen=2,
    path_policy="all_shortest",
    reuse_policy="induced_subgraph",
    only_signed=True,
    consensus=False,
)

The former algorithm, minimal, and connect_with_bias arguments are temporarily supported with a migration warning. algorithm="bfs" maps to one_shortest; algorithm="dfs" maps to all_bounded; minimal=True maps to discovered_paths; minimal=False with bias disabled maps to none; and either connect_with_bias=True combination maps to induced_subgraph. When no old or new selector is supplied, the transition release preserves the former effective default as all_bounded + discovered_paths without emitting a warning. Deterministic unweighted selection improves reproducibility but is not a biological ranking. Externally derived edge weights are planned for NeKo 2.0.

SIGNOR entity normalization

The built-in signor() input loads SIGNOR’s human interaction table and its complex, protein-family, phenotype, and stimulus dictionaries from NeKo’s validated local cache. Missing resources are downloaded once and added to the cache. Proprietary endpoint IDs are normalized before the Universe is built: complexes use the same COMPLEX: member syntax as OmniPath, while the other group/context nodes use readable PROTEIN_FAMILY:, PHENOTYPE:, and STIMULUS: identifiers.

from neko.inputs import signor

resources = signor()

After one successful load, the cached release can be used offline. Set NEKO_CACHE_DIR to choose the cache root. Preloaded dictionary DataFrames can still be passed through entity_dictionaries. Normalization can be explicitly disabled with normalize_entities=False when the raw SIGNOR identifiers are required.

SIGNOR ChEBI accessions remain canonical network identifiers and are never sent to UniProt for translation. When a resource contains ChEBI nodes, NeKo lazily downloads the official compressed compounds.tsv.gz table once and caches only the names needed by that resource for display. If the download is unavailable, network construction continues with the ChEBI accession as its label. ChEBI data are provided by EMBL-EBI under the Creative Commons Attribution 4.0 International license.

Installation

NeKo is distributed as Beta software. Install the nekomata distribution from PyPI; the Python import package remains neko.

1. Install NeKo from PyPI:

   Do not confuse nekomata with the unrelated neko or pyneko distributions.

   python -m pip install nekomata

Installation from Source

For the latest development version, you can still clone the repository and install directly from the source:

git clone https://github.com/sysbio-curie/Neko.git
cd Neko
pip install .

This will give you the latest version of NeKo (not officially released, so be aware there could be some bugs) along with the necessary external dependencies.

Troubleshooting

If Graphviz-related installation or rendering fails, install Graphviz using your system package manager.

sudo apt-get install python3-dev graphviz libgraphviz-dev

On macOS:

brew install graphviz

For more details visit: https://graphviz.org/download/

Documentation

For full documentation, including API reference and detailed tutorials, visit our GitHub Pages documentation. Users upgrading an existing workflow should also read the NeKo 1.9 migration guide.

Jupyter Notebooks

We provide a comprehensive set of Jupyter notebooks that offer a detailed and user-friendly explanation of the package. These notebooks cover all modules of NeKo and provide a complete overview of how to use the package:

1. Usage

2. Build network using user-defined resources

3. Stepwise connection: a focus on the INE algorithm

4. Connect to upstream components

5. Build network based on kinase-phosphosite interactions

6. Connect to downstream Gene Ontology terms

7. Map tissue expression

8. Network comparison

9. Re-creating famous pathways from SIGNOR and WIKIPATHWAYS using NeKo

10. Import and complete a network

11. Network history, branching, and visualisation

You can find these notebooks in the notebooks directory of the repository.

Features comparison with similar tools

Below you can find a table displaying the main features of NeKo compared to other similar tools: Features Table on GitHub.

Acknowledgements

This project is a collaborative effort between Institut Curie, NTNU, Saez lab and BSC.

Current contributors: Marco Ruscone, Eirini Tsirvouli, Andrea Checcoli, Dénes Turei, Aasmund Flobak, Emmanuel Barillot, Loredana Martignetti, Julio Saez-Rodriguez and Laurence Calzone.