Disease stratified network visualization

disease-stratified-network-visualization

Summary

Construct and render bipartite metabolite-protein association networks stratified by disease class, with visual encoding of association strength, confidence, and disease type. This skill transforms disease-specific correlation and prediction data into publication-ready network plots for systems-level interpretation of metabolic dysregulation.

When to use

You have metabolite-disease correlation scores and protein association predictions from a deep learning model (such as DeepMSProfiler) and need to visualize which metabolites and proteins co-associate within specific disease groups. Use this skill when stakeholders or reviewers require interpretable network-level evidence of disease-specific metabolic rewiring, especially when the number of associations is too large to present as tables or simple heatmaps.

When NOT to use

• Input data are already aggregated into a single network without disease stratification — use a standard network visualization tool instead.
• Metabolite or protein sets are too sparse (< 10 nodes per disease group) — consider alternative summaries such as ranked lists or heatmaps.
• Association scores lack confidence estimates or uncertainty quantification — network edge transparency cannot be meaningfully encoded.

Inputs

• metabolite-disease correlation matrix (numpy array or CSV: metabolites × diseases)
• protein association predictions (scores, confidence values per metabolite-protein pair)
• disease class labels (sample-level or metabolite-level annotations)
• association strength thresholds (optional: filtering cutoffs for edge inclusion)

Outputs

• bipartite network graph object (NetworkX or equivalent)
• network node layout coordinates (2D or 3D positions)
• high-resolution network plot image (PNG, SVG, or PDF suitable for publication)
• network graph file (optional: GraphML, GML, or JSON for external tools)

How to apply

Load metabolite-disease correlation data and protein predictions generated by the deep learning module. Construct a bipartite directed graph with metabolites and proteins as distinct node sets, filtering edges by association strength (e.g., correlation magnitude or prediction confidence above a threshold). Assign node size proportional to association magnitude and edge transparency or width proportional to confidence scores. Apply a force-directed or hierarchical layout algorithm to resolve node positions for visual clarity. Render the network with disease-type-specific color coding on nodes (e.g., one color per disease class), and export as high-resolution raster or vector format. Validate that nodes with high association strength cluster together and that disease-stratified coloring is visually distinct across groups.

Related tools

• DeepMSProfiler (source deep learning model that generates metabolite-disease correlations and protein association predictions) — https://github.com/yjdeng9/DeepMSProfiler
• NetworkX (Python library for constructing, manipulating, and analyzing bipartite graphs and applying layout algorithms)
• Cytoscape (optional interactive platform for post-hoc network refinement, filtering, and advanced visualization)

Examples

python showFeature.py  # Invokes feature extraction and network visualization pipeline post-training

Evaluation signals

• Network nodes are correctly stratified by disease class, visible via distinct color coding with no overlap in node identities across classes.
• Node sizes scale monotonically with association strength (magnitude of correlation or prediction confidence); verify by spot-checking high-magnitude and low-magnitude associations.
• Edge transparency or line width reflects confidence scores: high-confidence edges appear darker or thicker, low-confidence edges fainter or thinner.
• Force-directed layout produces visually coherent clusters with metabolites and proteins associated in the same disease group positioned nearby; hierarchical layout shows clear bipartite layering.
• Output image resolution is ≥300 dpi and file size is suitable for publication (< 50 MB for raster, < 10 MB for vector); visual distinctness of nodes, edges, and labels is confirmed at intended display size.

Limitations

• Network scalability: rendering becomes visually cluttered with > 500 nodes; consider filtering or hierarchical abstraction for larger association sets.
• Layout algorithm sensitivity: force-directed methods are stochastic; multiple runs may yield different (though topologically equivalent) layouts; set random seed for reproducibility.
• Disease stratification assumes discrete disease classes; continuous or multi-state phenotypes require custom node coloring or faceted subplots.
• No built-in statistical testing of network properties (e.g., clustering coefficient, shortest path significance) — these must be computed separately if required.
• File format compatibility: SVG and vector exports may not preserve all visual properties (transparency, gradients) when imported into all document processors.

Evidence

• [other] Construct a bipartite network graph with metabolites and proteins as nodes, weighted by association strength and disease class.: "Construct a bipartite network graph with metabolites and proteins as nodes, weighted by association strength and disease class."
• [other] Apply network layout algorithm (force-directed or hierarchical) to position nodes for interpretability.: "Apply network layout algorithm (force-directed or hierarchical) to position nodes for interpretability."
• [other] Render the network plot with disease-type color coding, node size scaled by association strength, and edge transparency reflecting confidence.: "Render the network plot with disease-type color coding, node size scaled by association strength, and edge transparency reflecting confidence."
• [other] Export the finalized network plot as a high-resolution image file suitable for publication.: "Export the finalized network plot as a high-resolution image file suitable for publication."
• [readme] Disease-associated metabolite-protein network plots. Load the metabolite-disease correlation data and protein association predictions generated by DeepMSProfiler's deep learning module.: "Load the metabolite-disease correlation data and protein association predictions generated by DeepMSProfiler's deep learning module."
• [readme] It takes raw metabolomics data from different disease groups as input and provides three main outputs: 1. Sample disease type labels. 2. Heatmaps depicting the correlation of different metabolite [signals with diseases]. 3. Disease-associated metabolite-protein network plots.: "It takes raw metabolomics data from different disease groups as input and provides three main outputs: 1. Sample disease type labels. 2. Heatmaps depicting the correlation of different metabolite"