Fig. 4

Network evaluation. The functional prediction performance of the low throughput (LTP_GS), BioSystems and Gene Ontology-scored networks as measured by area under curve (AUC) of receiver operator characteristic plots. The error of the AUC was calculated using the standard error of the Wilcoxon statistic SE(W): not statistically significant (NSS); statistically significant (SS). A. BioSystems-derived (MP_GS) Gold Standard. Of 398 Gene Ontology biological processes, 327 had improved prediction using the LTP_GS network, and 71 using the MP_GS network. Of these changes 227 were statistically significant with 209 (92%) improved using LTP_GS. B. Gene ontology biological process terms annotating <10% of the genome (BP10_GS) Gold Standard. Of 398 Gene Ontology biological processes, 230 had improved prediction using the LTP_GS network, compared to 168 using BP10_GS. Of the changes 114 were significant with 64 (56%) improved using LTP_GS. C. Gene ontology biological process terms annotating annotation <100 genes (BP100_GS) Gold Standard. Of 398 Gene Ontology biological processes, 105 had improved prediction using the LTP_GS network, compared to 293 using BP100_GS. Of the changes just 85 were significant with 3 (4%) improved using LTP_GS