Learning center

Graph assortativity: thresholds, undefined values and complete viewer exports

Read three synthetic graph shapes, degree mixing, explicit undefined outcomes and complete saved tables while preserving source, input and session limits.

Written guide to Assortativity in Computational Neuroscience Graph Metrics. Three synthetic graph shapes, exercised in four saved native runs, explain degree mixing and complete exports. One run renumbers the same star graph. These are analytical teaching fixtures, not recorded brains or an accepted clinical network model.

1. Declare what the matrix represents

Open Graph Metrics, identified as computational_neuroscience/Graph metrics. This guide uses a finite, symmetric connectivity matrix with zero diagonal and binary edges. Keep the node labels in the same order as its rows and columns. The saved four-node star connects node 0 to each other node:

0 1 1 1
1 0 0 0
1 0 0 0
1 0 0 0

The reviewed JSON input stores this matrix under connectivity, with channels containing node labels and units set to dimensionless. It also states an undirected matrix and an equal-weight threshold policy. Keep those declarations with the input. The production loader API was exercised; the full launcher Load/Validate/Apply sequence and its file chooser were not part of these receipts.

Shape changes the calculation: the app interprets a square upload as connectivity, while a rectangular upload becomes sample-by-node signals from which it derives correlations. A square time-series array can therefore be misinterpreted. For connectivity, the current source symmetrizes values, clears the diagonal and uses normalized absolute weights for thresholding. It does not preserve a directed or signed-network interpretation. Missing-value cleaning can interpolate values or drop entirely nonfinite columns; this example uses complete finite data and supplies no validation of those repairs.

2. Record the controls and the retained graph

Clear the twenty initially selected routes and retain Assortativity. The saved examples use Max nodes = 4 for the star/path or 3 for the triangle, Modules = 2, Edge density = 0.95 and Weight scale = 1.00. These are fixture settings, not recommended defaults for a connectome. The interface starts with Max nodes 36, Modules 4 and Edge density 0.18.

Max nodes is a selection limit. A larger square matrix is reduced to nodes ranked by absolute row-weight sum; a rectangular input is reduced by column variance. A smaller limit can change the graph, so retain the intended node set and compare the exported node count with the input.

The Edge density control ranks the currently positive edges. It retains weights at or above the selected cutoff, including ties; it does not guarantee that the final graph contains that fraction of all possible edges. In the equal-weight star, all three existing edges remain and the saved Density is 0.5, despite the control being 0.95. The triangle’s actual density is 1. Preserve the reported density as well as the control value.

3. Interpret the coefficient and explicit undefined values

Degree assortativity compares the degrees at the two ends of retained edges. This route uses binary degree after thresholding, not weighted strength. Each undirected edge contributes both endpoint orientations to the correlation. Positive values describe similar-degree connections; negative values describe different-degree connections. This is a dimensionless graph statistic, not a neural signal correlation, an anatomical location or a measure of health. The primary NetworkX documentation describes degree assortativity and gives the four-node path result.

GraphSaved rMeaning
Four-node star-1.0Each degree-3 hub endpoint connects to a degree-1 leaf endpoint.
Renumbered star-1.0The same graph shape with different node numbering retains the same value.
Four-node path-0.5Both end-to-middle and middle-to-middle connections contribute.
Three-node triangleundefinedEvery endpoint has degree 2, leaving zero degree variance for the correlation.

An undefined result is not zero and is not proof of neutral mixing. Retain AssortativityStatus, which explicitly identifies no edges or zero endpoint-degree variance. The triangle is a completed run with an undefined statistic. Separately, a native Run with no uploaded input was blocked and produced no Assortativity result; that input refusal is a different outcome.

4. Read all three figure panels

The Edge Degree-Degree Scatter includes both orientations. The star has three undirected edges and six endpoint pairs, although overlapping points can appear fewer. The Node Degree Profile shows degree by node index. An index is not an anatomical coordinate.

The Assortative Mixing Curve groups observed endpoint degrees and reports their mean neighbor degree. The star has only two groups: degree 1 maps to mean neighbor degree 3, and degree 3 maps to 1. The path has degree 1 mapping to 2 and degree 2 mapping to 1.5. Unobserved degree groups have no mean; a connecting line does not establish observations between the groups. The NetworkX average-degree-connectivity reference explains this grouped neighbor-degree quantity.

The triangle still has a valid observed group at degree 2 with mean neighbor degree 2, even though its correlation is undefined. A separate saved plotting check covers the no-edge explanation. That plotting check is not an additional native GUI run or a biological dataset.

5. Save the table container, selected figure and viewer session

The native viewer has distinct actions for exporting tables, exporting the selected plot and saving a session. Each fixture’s reviewed set contains three separately saved files:

Saved outputWhat it retains
native-tables.csvAll seven viewer tables in a Table,Metric,Value container: 50 data rows, including all ten Assortativity fields.
native-plot.pngThe selected three-panel Assortativity figure at 2304 × 768 pixels. It is not an export of every viewer figure.
native-session.jsonTwo plot snapshots, seven tables, five text reports and viewer state. The prior native receipt reopened this file in the viewer.

The table container includes Input Contract, QC Validation, Domain Contract, Domain QC, Loaded Data Context, Assortativity and Run Summary. The app’s run manifest counts six tables before the viewer adds Run Summary; the saved viewer CSV/session contains seven. Select rows by their Table field instead of treating all 50 rows as one numerical matrix.

The guide checks all 200 CSV fields against the four saved sessions and all forty Assortativity fields against the native receipt. It also verifies the twelve file identities. The numeric coefficient remains numeric in the saved session; undefined remains explicit text. Other summary values, such as MeanDegree and MeanStrength, were formatted to six significant digits before export. Exporting them does not recover extra precision.

The session contains raster snapshots and embedded SVG/PDF representations for the two plots. Their saved byte identities were checked without a new rendering or analysis. Reopening a viewer session is not the same as restoring the original app’s loaded scientific input and rerunning its calculation. Retain the JSON matrix, settings, run manifest and evidence hashes separately. Inspect the saved paths and provenance before sharing; this three-file set has no separate native artifact manifest.

6. Keep QC warnings and scientific limits visible

The saved fixtures complete with QC and Domain QC status “warn”, with no listed failures. Missing temporal, spatial and event/trial metadata remains explicit. The examples support non-temporal, non-spatial graph interpretation; they do not establish anatomical, behavioral or clinical claims. The generic Loaded Data Context panel’s sample/channel labels do not turn this adjacency matrix into a sampled recording.

The native checks establish the stated small graph examples and export behavior. They do not establish a suitable threshold for real data, signed or directed assortativity, every other Graph Metrics route, large-network performance, hosted execution or biological validity. A repeated or renumbered fixture does not become a new independent dataset.

Worked examples describe particular inputs, methods and recorded results. Check current availability and input requirements before planning an evaluation.

← Back to the learning center