Skip to main content

Table 3 Methods used to assess co-expression networks

From: Exploiting single-cell expression to characterize co-expression replicability

Assessment Property tested Method and interpretation
Co-expression performance Functional connectivity of each GO group • Neighbor voting for GO functions, threefold cross-validation
• AUROC >0.5 means the genes within the GO group have greater connectivity than may be expected by chance
Within-network semantic similarity (thresholded) Functional connectivity of the network • Number of GO functions in common among top 1 % of network connections
• A higher value indicates greater functional similarity among top connections
Between-network similarity (thresholded) Topology • Jaccard index (intersect/union) over top 1 % connections
• Values closer to 1 imply greater similarity between networks
Co-expression performance after aggregation Replicability of functional connectivity • Neighbor voting for GO functions as datasets are added together to form aggregate networks
• Improvement with aggregation suggests specific replicability of functional connections
Aggregate connectivity distributions Replicability of topology • Rank standardized networks (with uniform distributions) are added together to form an aggregate network, then the standard deviation across all co-expression values is measured
• Higher variance indicates greater replicability of topology among networks