Skip to main content
. 2011 Feb 18;6(2):e17258. doi: 10.1371/journal.pone.0017258

Figure 6. Potential node degree corrections do not remove node degree influence.

Figure 6

A) The unsparse coexpression data exhibits low node degree bias and very good performance using a nearest neighbor voting algorithm (mean AUC of 0.71). B) A very strong dependency on node degree is retained in the nearest neighbor analysis. The more extreme the performance of a GO group is based on node degree, the better the performance in the nearest-neighbor analysis. C) After a correction to remove the influence of node degree – links included only if performance is above that predicted by node degree alone – and aggregation, this matrix exhibits high performance (mean AUC = 0.70) and modest node degree bias (mean AUC = 0.55), but it again exhibits the triangular dependence on multifunctionality. D) An unsparsified coexpression network from a large multi-tissue coexpression experiment (GSE7307) yields a mean AUC of 0.68 across all GO groups and no node degree bias in favor of multifunctionality (mean AUC = 0.48), but exhibits the same triangular distribution of node degree AUCs.