Random Walk with Restart over 13 evidence networks (PPI, pathways, co-expression, function,
phenotype, mouse models, regulation, complex, co-essentiality, drug, co-localization,
genomic, co-citation), one per knowledge category. Per-gene scores are weighted by
disease-aware recall × specificity
(cross-validated on the seed set).
Candidates outside the seed set with high score are plausible novel disease-gene predictions.
Method: de la Fuente et al.,
Int. J. Mol. Sci.
2023, 24:1661
[doi]
—
GLOWgenes repo
. Networks (CC BY-NC-SA): GLOWgenesNets v1 (figshare 21408393).
Networks are pre-loaded at app startup. Set CV = 0 to skip cross-validation (uniform weights, faster but less informative).
PanelApp panels most loaded with our top-N predictions
For each Genomics England PanelApp panel, the bar shows how many of our live top-
Top-N
candidates are also in that panel's top-N pre-computed by the
GLOWgenes authors. Top-of-list panels suggest which clinical entities your phenotype is
most aligned with — independent of which seeds you started from.
PanelApp panel — pre-computed ranking
Pick any panel above (or below) to inspect the published GLOWgenes ranking for
that disease as-is, plus how those genes rank in our live computation.
Rank vs RWR score (coloured by PanelApp context)
X = rank (1 = best, on the left), Y = RWR score. Colour:
black
seed gene from your input;
red
in the top-N of the panel picked above;
blue
known to at least one panel;
grey
novel candidate. Hover for top supporting network.
Predicted novel candidates (live ranking)
Genes ranked by combined RWR score across the 13 evidence networks.
Your seed genes are excluded from this list
— every row is a network-proximal
candidate that is
not
in your input set.
panelapp_seed_count
= number of PanelApp panels in which this gene is a
canonical (rank-0) seed → high values mean a known pleiotropic disease gene; zero means
a fully novel network-derived prediction.
Download full ranking
Method-level diagnostics, not clinical outputs.
Per-network weights
show which knowledge category drove the ranking
(networks with recall = 0 contributed nothing).
Top genes × network heatmap
shows whether each top candidate is supported
by many networks (robust) or just one (potentially spurious).
Per-network weights (disease-aware)
Top genes × network heatmap (raw RWR scores)
STRING physical PPI v12 (experimentally observed, no text-mining) on the primary set
+ top interactors. Nodes in
blue (orange label)
= seeds;
grey
= interactors.
STRING score
∈ [0, 1000] = interaction confidence.
Blue
edges = seed-seed;
grey
= seed-neighbor; thickness ∝ score.
Type:
seed-seed
= edge between two input genes (internal to the module);
seed-neighbor
= edge between an input gene and an external interactor (potential novel candidate).
For each gene in the set, its top 30 partners by expression-profile
correlation across the ~70 GTEx tissues. Column
in_hpo_universe
indicates whether the partner already has HPO annotation (those that
do NOT are potential novel candidates).
Same logic as GTEx, but the correlation is computed across the ~74
Tabula Sapiens tissues. Different sample structure surfaces partners
that GTEx may miss (retina, cornea, cochlea cell types, etc.).
