Network Visualization

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

Audit 0.4.1 · source sha c8c1fc3a2b13 · 19 Oct 2026Report an issue or request a re-scan

• PublisherGPTomics GitHub organizationOwner of the repository this skill is read from
• Source folderGPTomics/bioSkills/data-visualization/network-visualizationThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdictc8c1fc3a2b13A new audit runs when the source changes
• Repository statusArchivedLast push 65 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Visualize biological networks (PPI, gene-regulatory, co-expression, pathway) with layout algorithm choice (ForceAtlas2, Fruchterman-Reingold, Kamada-Kawai, hive plots), edge bundling, community-based coloring, and…

---
name: bio-data-visualization-network-visualization
description: Visualize biological networks (PPI, gene-regulatory, co-expression, pathway) with layout algorithm choice (ForceAtlas2, Fruchterman-Reingold, Kamada-Kawai, hive plots), edge bundling, community-based coloring, and reproducible seeds using NetworkX, PyVis, igraph, and Cytoscape automation. Use when rendering biological networks for static publication, interactive HTML exploration, or Cytoscape-format export.
tool_type: python
primary_tool: NetworkX
---

## Version Compatibility

Reference examples tested with: networkx 3.2+, igraph 0.10+ (Python and R), pyvis 0.3+, py4cytoscape 1.9+, matplotlib 3.8+, datashader 0.16+ (for large-graph rasterization).

Before using code patterns, verify installed versions match. If versions differ:
- Python: `pip show <package>` then `help(module.function)` to check signatures
- R: `packageVersion('<pkg>')` then `?function_name`

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

# Network Visualization

**"Plot a biological network"** -> Select a layout algorithm (force-directed for general; hive plot for comparative; ForceAtlas2 for scale-free; circular for small dense), encode node attributes (size by degree/centrality, color by community/module), and choose rendering tier (matplotlib for static publication; PyVis for interactive HTML; Cytoscape for journal-grade compositing). The dominant pitfall is treating layout as biology — node positions in force-directed plots are NOT biologically meaningful; only connectivity is.

- Python: `networkx`, `pyvis.Network`, `py4cytoscape`, `datashader` (large graphs)
- R: `igraph`, `ggraph` (ggplot2-grammar for networks)
- Desktop: Cytoscape (Shannon 2003), Gephi (ForceAtlas2 native)

## The Single Most Important Modern Insight -- Layout Is an Artifact, Not Biology

A force-directed layout (Fruchterman-Reingold, ForceAtlas2, spring) is the result of an optimization that minimizes edge crossing and balances repulsion. The visual position of a node has no biological meaning — it is determined by the layout algorithm + random initialization + iteration count + repulsion parameters.

Two consequences:
1. **Set `random_state` / `seed` for reproducibility.** Without it, the same network produces different layouts across runs.
2. **Do not read "cluster A is closer to cluster B than C" as biology.** Inter-community distances in force-directed layouts are not preserved. Only EDGE existence and node DEGREE are biological signals from the visual.

For biology-faithful layouts, use **hive plots** (Krzywinski 2012) which anchor nodes to fixed axes by metadata, OR **circular** layouts which preserve symmetry but don't claim distance meaning.

## Decision Tree by Network Type and Question

| Network | Recommended layout | Reason |
|---------|--------------------|--------|
| Generic PPI (<500 nodes) | Fruchterman-Reingold OR Kamada-Kawai | General-purpose; clean separation |
| Scale-free PPI (>500 nodes, hub-spoke) | ForceAtlas2 (Jacomy 2014) | Designed for scale-free networks |
| Gene regulatory (directed) | Hierarchical OR ForceAtlas2 with edge direction | Direction matters; hierarchical for cascade |
| Pathway / signaling | Manual or Cytoscape layout | Curated layouts in WikiPathways/Reactome |
| Co-expression module visualization | Hive plot anchored by module assignment | Comparative; nodes by category |
| Many-to-many (>10k edges) | Hierarchical edge bundling (Holten 2006) | Reduces visual clutter |
| Large network (>50k nodes) | Datashader raster + interactive zoom | matplotlib chokes; raster is the only honest display |
| Connectivity-only (no positions) | Adjacency matrix heatmap | Network as matrix avoids layout artifact |
| Comparing two networks | Side-by-side same layout (`pos` reused) | Otherwise layout differences mask biology |

## Layout Algorithms

```python
import networkx as nx

# Spring / Fruchterman-Reingold (general)
pos = nx.spring_layout(G, k=1/np.sqrt(len(G)), iterations=100, seed=42)

# Kamada-Kawai (better for small dense)
pos = nx.kamada_kawai_layout(G)

# Circular
pos = nx.circular_layout(G)

# Shell (hub at center, periphery outside)
pos = nx.shell_layout(G, nlist=[hub_nodes, periphery_nodes])

# Spectral (reveals clusters)
pos = nx.spectral_layout(G)

# Bipartite (two sets)
pos = nx.bipartite_layout(G, top_nodes)

# Hierarchical (DAG)
pos = nx.nx_pydot.graphviz_layout(G, prog='dot')   # requires graphviz
```

For ForceAtlas2 in Python: `fa2_modified` (newer maintained fork) or use Gephi for the canonical implementation. For ggraph in R:

```r
library(ggraph)
ggraph(g, layout = 'fr') +                          # Fruchterman-Reingold
    geom_edge_link(alpha = 0.3) +
    geom_node_point()

ggraph(g, layout = 'kk') +                          # Kamada-Kawai
ggraph(g, layout = 'circle') +
ggraph(g, layout = 'graphopt') +                    # OpenOrd-style for large
```

## Hive Plots (Krzywinski 2012) — Biology-Faithful

A hive plot anchors nodes to 2-3 fixed axes by a categorical attribute (e.g., node type, module, chromosome); edges drawn as arcs between axes. Removes the "hairball" effect by replacing free 2D layout with structured 1D axes.

```python
# HiveNetX or pyveplot for hive layouts
# Or use d3.js HivePlot for interactive
# R: HivePlotData via igraph + custom rendering
```

Use hive plots when comparing networks across conditions OR when nodes have a categorical structure (e.g., TFs vs targets, chromosomes for 3D-genome interactions).

## Hierarchical Edge Bundling (Holten 2006)

For many-to-many networks within a hierarchical structure (gene hierarchies, taxonomies), edge bundling routes edges along the tree backbone, dramatically reducing clutter.

```r
library(ggraph)
ggraph(graph, layout = 'dendrogram', circular = TRUE) +
    geom_conn_bundle(data = get_con(from = from_idx, to = to_idx),
                     alpha = 0.

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

pip install networkx pyvis py4cytoscape matplotlib numpy
pip install datashader

git clone https://github.com/GPTomics/bioSkills
git -C bioSkills checkout c8c1fc3a2b13

Works in

1. 2026-10-19
   ✓SAFE B 89 / 100

   first audit · source c8c1fc3a2b13

Questions about Network Visualization

• github.com/GPTomics/bioSkills/tree/main/data-visualization/netwo — the audited source at commit c8c1fc3a2b13 and every finding aboveObserved · 19 Oct 2026

Spotted something wrong? Report a correction · Methodology