KAN-GRN

Gene Regulatory Network Inference using Kolmogorov-Arnold Networks

KAN-GRN infers Gene Regulatory Networks (GRNs) from single-cell RNA sequencing data using Kolmogorov-Arnold Networks (KANs).

Installation

pip install kan-grn

For GPU support:

pip install torch --index-url https://download.pytorch.org/whl/cu124
pip install kan-grn

Quick Start

Command Line

# Basic usage
kan-grn run expression_data.h5ad network_file.tsv

# With custom parameters
kan-grn run expression_data.h5ad network_file.tsv --output-dir results --n-top-genes 3000 --epochs 100 --device cuda

Python API

import kan_grn

pipeline_config = kan_grn.PipelineConfig(
    expression_file="expression_data.h5ad",
    network_file="network.tsv",
    output_dir="results",
    n_top_genes=3000,
    device="cuda",
    model_config=kan_grn.ModelConfig(grid=5, k=4),
    training_config=kan_grn.TrainingConfig(
        batch_size=128,
        epochs=100,
        learning_rate=0.001,
        generate_symbolic=True,
    ),
)

pipeline = kan_grn.KANGRNPipeline(pipeline_config)
results = pipeline.train_models_only()

CLI Options

Input Formats

• Expression data: .h5ad (AnnData) with cells × genes matrix
• Network file: .tsv with columns: source_gene, target_gene, weight

Requirements

Python ≥ 3.8, PyTorch ≥ 1.9.0, Scanpy ≥ 1.9.0, PyKAN

Links

• GitHub Repository
• Documentation
• Issues

License

GPL-3.0

Author

Posh Raj Dahal (dahal.poshraj24@gmail.com)