PyNetworKIN

PyNetworKIN is a Bayesian kinase–substrate prediction pipeline for phosphoproteomics. It integrates sequence-motif scoring (via pynetphorest) with protein-interaction context (via the STRING network) to predict which kinases, phosphatases, or phospho-binding domains are responsible for observed phosphorylation events.

This repository is a modernised Python 3 port of the original NetworKIN 3.0 tool (Linding, Jensen, Horn & Kim, 2005–2013), extended to support STRING v12 protein interaction data.

Features

• Predicts kinase/phosphatase/phospho-binding domain substrates from FASTA + phosphosite input.
• Supports human (9606) and yeast (4932) proteomes.
• Accepts multiple phosphosite input formats: NetworKIN TSV, ProteomeDiscoverer, MaxQuant, and custom formats.
• Integrates sequence motif posterior probabilities with STRING network proximity scores using pre-calibrated Bayesian likelihood-ratio tables.
• Outputs per-site predictions as a TSV file in the results/ directory.

Requirements

Installation

From source

pip install -e .

Docker (GHCR)

docker pull ghcr.io/bibymaths/pynetworkin:latest
docker run --rm -v "$(pwd):/work" ghcr.io/bibymaths/pynetworkin:latest predict /work/input.fasta

Or use the provided Compose file:

docker compose up -d
docker compose exec networkin pynetworkin predict /work/input.fasta

Usage

CLI

pynetworkin predict <FASTA-file> [options]

Example

pynetworkin predict data_MaxQuant_sample/test.fasta --output results/test.networkin.tsv

Results are written to results/<fasta-filename>.result.tsv.

Other commands

pynetworkin info       # Show runtime/package information
pynetworkin cache      # Show cache contents
pynetworkin cache --clear  # Clear cached network data

Python API

from pynetworkin import AppConfig, run_pipeline

config = AppConfig(
    organism="9606",
    fasta_path="data_MaxQuant_sample/test.fasta",
    sites_path=None,
    datadir="data",
    blast_dir="",
)
results = run_pipeline(config)
print(results["prediction_count"], "predictions written to", results["output_path"])

Input formats

FASTA file

Standard FASTA format. Protein IDs are taken as everything between > and the first _ on the header line.

Sites file (auto-detected)

Output format

Results TSV columns:

Repository structure

src/
  pynetworkin/          # Core pipeline package
    __init__.py         # Public API (AppConfig, run_pipeline)
    networkin.py        # Main pipeline: AppConfig, run_pipeline, detect_site_file_type, …
    motif_scoring.py    # pynetphorest batch scorer wrapper
    graph_scoring.py    # STRING network context scoring & prediction ranking
    likelihood.py       # Bayesian likelihood conversion tables
    logger.py           # Loguru/Rich logging wrapper
    output.py           # TSV / Cytoscape SIF output writers
    recovery.py         # False-negative recovery via network proximity
    cli.py              # Typer CLI entry-point
    inputs/
      phosphosites.py   # OmniPath / PhosphoSitePlus / fallback fetcher
      string_network.py # STRING flat-file / REST API / fallback fetcher
scripts/
  backup.py                  # Legacy NetworKIN 3.0 reference script (Python 3 port)
  cleanup_HGNC_mapping.py    # HGNC symbol–Ensembl ID reconciliation utility
  generate_sample_data.py    # Generate offline fallback data files
  migrate_to_parquet.py      # Migrate legacy .txt conversion tables → Parquet
data/
  conversion_direct.parquet   # Pre-built likelihood tables (direct STRING paths)
  conversion_indirect.parquet # Pre-built likelihood tables (indirect STRING paths)
  fallback/                   # Bundled offline sample data
  string_data/                # STRING interaction flat files
tests/
  conftest.py             # pytest path setup (adds src/ to sys.path)
  test_motif_scoring.py
  test_output.py
  test_recovery.py
  test_networkin.py       # Tests for load_conversion_tables, detect_site_file_type, run_pipeline

See ARCHITECTURE.md for a detailed description of the execution flow.

Data sources

• pynetphorest: kinase-group motif models (Python package).
• STRING v12: human protein interactions and sequences. Downloaded from string-db.org.
• OmniPath: phosphorylation site reference data (fetched live, cached locally).

This repository provides a modern reimplementation of the NetworKIN framework.

• Original NetworKIN was described in: Linding et al., Cell 2007

• This implementation:

  • Does NOT reuse original NetworKIN source code
  • Replaces NetPhorest with pynetphorest
  • Uses a rewritten likelihood model
  • Implements a new modular pipeline

License: MIT