mossn 0.1.0

MOSSN: sample-specific protein network inference from gene expression and multi-omics data

mossn

mossn packages the MOSSN algorithm for constructing sample-specific protein interaction networks from gene expression data, together with ablation variants and multi-omics extensions.

Features

• Sample-specific edge reweighting using gene-expression-derived correction scores.
• Random walk with restart (RWR) to estimate node importance per sample.
• Ablation variants:
  • no_prior
  • uniform
  • no_seed
  • no_rwr
  • no_corr
• Multi-omics extensions:
  • coupled restart
  • direct cross-layer graph
  • multilayer graph
  • late fusion

Installation

pip install mossn

For local development:

pip install -e .

Quick Start

from mossn import prepare_data_no_prior, run_no_prior_single_sample
from mossn.example_data import load_example_expression, load_example_links

links = load_example_links()
expression_data = load_example_expression()

graph, base_weights, expression_data = prepare_data_no_prior(links, expression_data)
edge_table = run_no_prior_single_sample(
    sample_id=expression_data.columns[0],
    graph=graph,
    base_weights=base_weights,
    expression_data=expression_data,
)

print(edge_table.head())

Bundled Example Data

The package includes the following example datasets:

• TCGA BLCA expression matrix
• STRING-derived PPI links

You can access them with:

from mossn.example_data import (
    get_example_expression_path,
    get_example_links_path,
    load_example_expression,
    load_example_links,
)

Input format

PPI links

The links table must contain:

• protein1
• protein2
• score

Expression matrix

The expression matrix must use:

• rows as genes or proteins
• columns as sample IDs

Main API

Single-omics

• prepare_data_no_prior
• run_no_prior_single_sample
• prepare_data_uniform
• run_uniform_single_sample
• prepare_data_no_seed
• run_no_seed_single_sample
• prepare_data_no_rwr
• run_no_rwr_single_sample
• prepare_data_no_corr
• run_no_corr_single_sample

Data-driven extension

• prepare_data_driven
• build_graph_from_correlation
• run_driven_single_sample
• infer_driven_network

Multi-omics

• prepare_data_coupled
• run_coupled_single_sample
• prepare_data_direct
• run_direct_single_sample
• prepare_data_multilayer
• run_multilayer_single_sample
• run_late_fusion_single_sample

Notes

• The package expects matched identifiers between the network and omics tables.
• Sample-specific normalization uses median and interquartile range (IQR).
• Node importance is rank-normalized before computing final edge weights.
• The data-driven mode first infers a background graph from expression correlations when an external reference network is unavailable.

Data-Driven Example

from mossn import infer_driven_network, run_driven_single_sample
from mossn.example_data import load_example_expression

expression_data = load_example_expression()
graph, base_weights, expression_data = infer_driven_network(
    expression_data=expression_data,
    cor_threshold=0.9,
)

edge_table = run_driven_single_sample(
    sample_id=expression_data.columns[0],
    graph=graph,
    base_weights=base_weights,
    expression_data=expression_data,
)